Wearable Sensors for Smart Agriculture: From Sensing Mechanisms to Crop Phenotyping and Harvesting Applications

Abstract

Global food security faces severe pressures from population growth, climate change and resource shortages. Traditional agricultural monitoring relies on manual sampling and offline laboratory testing, which involve considerable time delays and cannot realize continuous, in situ dynamic tracking of crop physiological status, field environment and agricultural product quality. Among the various sensing platforms for agriculture, flexible wearable sensors represent an important branch, featuring light weight, conformability, low power consumption and real-time response. It should be noted that rigid-substrate and silicon-based sensors remain indispensable in many agricultural applications, and this review focuses specifically on flexible wearable sensors while acknowledging the broader landscape of agricultural sensing technologies. Through conformal contact with plant surfaces or wearable integration, they can convert physical, chemical and biological signals of agroecosystems into readable electrical/optical signals via specific sensing transduction mechanisms. This review systematically categorizes wearable sensors for smart agriculture into two main types, optical and electrical, and elaborates their intrinsic sensing principles, summarizes typical application scenarios covering crop growth monitoring, post-harvest harvesting and agricultural product safety detection, and introduces the full-flow signal preprocessing, feature extraction and intelligent analysis technologies matched with wearable sensing data. Finally, the bottlenecks restricting large-scale field deployment of current wearable sensors are summarized, and the development directions of multi-modal integrated wearable sensing, biomimetic nanosensors inspired by natural plant sensing systems, and artificial intelligence (AI)-assisted agricultural monitoring platforms are prospected, aiming to provide a theoretical reference for the design optimization and popularization of agricultural wearable sensing equipment.

1. Introduction

Global agricultural systems face unprecedented challenges driven by climate variability [1], arable land degradation [2], freshwater scarcity [3], and an exponentially growing world population [4,5,6,7]. Ensuring global food security while minimizing ecological footprints demands an urgent paradigm shift from traditional empirical farming to precision and smart agriculture [8,9,10,11]. A cornerstone of this technological transformation is the continuous acquisition of physiological and biochemical datasets from both crop-soil microenvironments and post-harvest produce, with the required spatial and temporal resolution depending on the specific agricultural application [12,13,14,15]. Historically, agricultural sensing and phenotyping have progressed through four representative technological phases (Figure 1): (i) manual sampling coupled with destructive, offline laboratory assays (e.g., high-performance liquid chromatography and mass spectrometry); (ii) discrete in-field testing with handheld electroanalytical or optical meters; (iii) stationary, rigid agro-meteorological monitoring nodes and remote sensing platforms (e.g., UAV-based multispectral imaging) [16,17,18]; and (iv) conformal, bio-integrated wearable sensors enabling non-invasive, continuous, and in situ dynamic profiling [19,20]. These phases should not be regarded as a strict hierarchy of technological superiority, but rather as successive and partially overlapping stages in research, development, and application. Each approach addresses different agricultural monitoring needs and has its own suitable application scenarios. For example, remote sensing is advantageous for large-scale canopy monitoring, benchtop analytical methods remain indispensable for high-accuracy laboratory quantification, and non-wearable or plant-integrated sensors are often more suitable for specific environmental or soil measurements. Wearable sensors are therefore not a universal solution, but a complementary tool for applications that require direct, continuous, and in situ monitoring of plant physiological and biochemical signals. While remote sensing offers macro-scale canopy coverage, it inherently lacks single-organ spatial resolution and early molecular-level sensitivity [21]. Among remote sensing platforms, drones (unmanned aerial vehicles, UAVs) have become an important tool for crop detection because they can provide flexible, low-cost, and high-throughput monitoring at field scale. Equipped with multispectral, hyperspectral, RGB, thermal, or LiDAR sensors, drones can rapidly map crop growth status, canopy structure, chlorophyll content, water stress, and disease incidence over large areas, and they are particularly useful for precision irrigation, variable fertilization, and early stress screening. However, drone-based crop detection still operates mainly at the canopy or plot level, and its spatial resolution is generally insufficient for single-organ or single-plant physiological profiling. Moreover, drone remote sensing is susceptible to weather conditions, illumination variation, and cloud cover, and it cannot directly measure internal biochemical signals such as phytohormones, ROS, or sap ion concentrations. Therefore, drones and plant-wearable sensors are complementary rather than substitutive: drones provide macro-scale, high-throughput canopy information, while wearable sensors provide micro-scale, continuous, and in situ physiological and biochemical data. Integrating drone remote sensing with ground-based wearable sensing is expected to enable multi-scale crop monitoring from canopy to organ level. Conversely, traditional benchtop analytical methods suffer from substantial time latency, high operational costs, and the inability to capture transient chemical signaling pathways under dynamic field stresses. These limitations create a clear niche for conformal wearable sensors that can provide direct, continuous, and in situ monitoring at the plant-organ level. Given the diversity of sensing mechanisms and application targets, a basic overview is helpful before the detailed discussion. Accordingly, the main categories, sensing principles, and typical agricultural applications of wearable sensors are summarized in Table 1.

Figure 1. Evolution roadmap of agricultural sensing monitoring technology. Source: Authors’ own illustration.

Table 1. Basic classification and sensing principles of wearable sensors for smart agriculture.

Sensor Category	Main Subtype	Typical Sensing Principle	Typical Measured Signals	Typical Agricultural Applications
Optical	Luminescence (fluorescence, bioluminescence, chemiluminescence, etc.)	Fluorescence-based mechanisms: FRET, PET, fluorescence quenching	Fluorescence intensity, emission wavelength	ROS, phytohormones, pesticide residues, disease biomarkers
Optical	SERS	Localized surface plasmon resonance, electromagnetic/chemical enhancement	Raman shift, Raman intensity	Pesticide residues, mycotoxins, pathogen secretions
Electrical	Electrochemical	Amperometric/voltammetric, potentiometric, impedimetric transduction	Current, potential, impedance	Sap nutrients, ions, phytohormones, water status
Electrical	Piezoresistive	Percolation and quantum tunneling, strain-dependent resistance	Resistance change	Stem/leaf growth strain, drought stress, grasping force
Electrical	Capacitive	Dielectric permittivity change, electrical double layer	Capacitance change	Transpiration, humidity, pressure, contact sensing

Plant wearable sensors [22], emerging at the intersection of flexible electronics, advanced nanomaterials chemistry, and plant biology, have unlocked transformative pathways to circumvent these classical bottlenecks [23,24,25]. Theoretically, their sensing performance is governed by interfacial charge transfer, energy transfer, and mechanical strain transduction at the sensor plant interface. In terms of device implementation, these devices typically employ functional nanocomposites (e.g., transition metal dichalcogenides [26,27], MXenes [28], conducting polymers [29], and quantum dots [30,31]) integrated with biocompatible flexible substrates [32] (such as PDMS, cellulose, and natural hydrogels [33]) [34,35,36,37]. In practical production, wearable devices can achieve non-destructive long-term adhesion on crop tissues without damaging plant growth, and complete in situ trace detection of stress biomarkers and pesticide residues, as well as ripeness assessment of fruits, effectively reducing the loss caused by crop diseases and improper harvesting [38,39,40]. From a practical perspective, flexible wearables enable non-destructive, long-term adhesion to micro-curved and delicate plant surfaces [41] (e.g., abaxial leaf stomata, expanding stems, and ripening fruit pericarps) [42]. This conformal bio-integration facilitates in situ trace quantification of stress phytohormones (abscisic acid, jasmonic acid, and salicylic acid), reactive oxygen species (ROS) [43,44,45], secondary metabolites, pesticide residues, and biomechanical strain with picomolar-to-micromolar sensitivity, providing an actionable diagnostic window prior to the onset of macroscopic phenotypic symptoms.

Recent pioneering efforts have accelerated the design of diverse plant-wearable modalities [46,47,48]. For instance, plasmonic surface-enhanced Raman scattering (SERS) tapes and aptamer-functionalized fluorescent patches have been engineered to achieve label-free chemical fingerprinting of pesticide sprays and real-time ROS tracking under acute drought and thermal shock [46,49,50]. Concurrently, microfabricated electrochemical microneedles and laser-induced graphene transducers have demonstrated stable potentiometric/amperometric monitoring of sap nutrients, ionic homeostasis (K⁺/Na⁺/NO₃⁻), and stomatal conductance dynamics [51,52,53,54]. Furthermore, piezoresistive and capacitive flexible strain sensors have been widely deployed on fruit stalks and robotic end-effectors to decipher diurnal micro-variations in stem expansion and prevent mechanical damage during automated harvesting [55,56,57].

Despite these significant strides, the widespread agricultural translation of wearable sensors remains severely constrained by several fundamental challenges [58,59,60]: (1) Modality limitation: Most existing sensors operate on single-parameter transduction, lacking multi-modal capability to synchronously decouple coupled environmental and physiological inputs (e.g., co-monitoring foliar transpiration, local microclimate, and systemic hormone fluxes) [61]. (2) Severe environmental degradation: In harsh open-field conditions, sensor performance rapidly deteriorates due to baseline drift and biofouling induced by cyclic ultraviolet radiation, agrochemical spray erosion, ambient humidity fluctuations, and airborne particulate accumulation. (3) Complex bio-matrix interferences: Plant cuticular waxes, background autofluorescence, endogenous redox-active metabolites, and non-target co-existing ions frequently compromise the selectivity, signal-to-noise ratio, and operational lifespan of sensitive layers [62]. (4) Data processing bottlenecks: Continuous in situ data acquisition generates voluminous time-series streams plagued by environmental noise and motion artifacts; however, dedicated agronomic chemometric pipelines and lightweight edge-computing algorithms for real-time drift calibration and stress-event classification remain underdeveloped [63]. (5) Scalability and cost-efficiency: High-performance nanofabrication routes frequently depend on high-vacuum or cleanroom lithography processes, which are economically incompatible with large-scale, distributed deployment in practical farming systems [64].

To bridge these knowledge and technological gaps, this review provides a comprehensive and critical overview of agricultural wearable sensors, linking molecular-level transduction chemistry to smart farming practices and post-harvest safety assurance. We first categorize the fundamental mechanisms of optical (fluorescence, FRET, and SERS) and electrical (electrochemical, piezoresistive, and capacitive) wearable sensors, formulating quantitative mathematical models that govern their interfacial response behavior. We then delve into dual application chains: (i) whole-lifecycle crop growth and stress biomarker phenotyping (ROS, phytohormones, water deficit, and phytopathogens), and (ii) intelligent harvesting, grasping control, and rapid agro-product quality/safety screening (pesticide residues, fungal mycotoxins, and ripeness indicators). Subsequently, the end-to-end signal processing framework, encompassing noise filtering, matrix calibration, multivariate chemometrics, and machine-learning-assisted edge/cloud diagnostics, is systematically evaluated. Finally, we discuss key prospective directions, highlighting multimodal integrated micro-systems, self-powered bio-triboelectric/piezoelectric energy harvesters, anti-fouling interfacial coatings, and roll-to-roll scalable manufacturing, thereby providing a multidisciplinary roadmap for next-generation agricultural monitoring technologies.

2. Classification and Sensing Principles of Wearable Sensors

Flexible wearable sensors represent a paradigm-shifting diagnostic modality in precision agriculture, capitalizing on low mechanical modulus, conformal interfacial adherence, and biocompatible functionalization to seamlessly interface with soft, curvilinear, and dynamic plant tissues (e.g., stomata-bearing abaxial foliar surfaces, elongating stems, expanding xylem/phloem vessels, and ripening pericarps) [65,66]. Based on their intrinsic signal transduction mechanisms and physicochemical interfacial behaviors, agricultural wearable sensors are primarily categorized into optical transducers and electrical transducers. Optical modalities transduce molecular recognition events and microenvironmental perturbations into modulations of radiative photonic states, absorption cross-sections, or inelastic Raman scattering [67,68,69]. In contrast, electrical modalities convert interfacial biochemical redox reactions, potential gradients, mobile ion fluxes, or mechanical strains into readable amperometric, potentiometric, resistive, or capacitive signals. This section systematically dissects the theoretical foundations, interfacial mechanics, mathematical formulations, and operational physics underpinning both categories of plant-wearable sensors [70,71,72].

2.1. Fundamental Principles of Optical Wearable Sensors

Optical plant wearables are typically constructed on elastic polymeric substrates (e.g., polydimethylsiloxane [PDMS], nanofibrillated cellulose, transparent polyimide, silk fibroin) functionalized with photonic nanomaterials, including colloidal quantum dots (QDs) [73,74,75], carbon dots (CDs) [76], lanthanide-doped upconversion nanoparticles (UCNPs) [77,78], metal–organic frameworks (MOFs) [79,80], and plasmonic nanoarrays [81,82,83,84]. By probing photonic state transitions, optical sensors enable non-invasive, remote, and electromagnetic-interference-free dynamic tracking without demanding hardwired electrical interconnects across fragile vegetative organs [85,86] (Figure 2).

Figure 2. Schematic illustration of typical optical sensing nanoplatforms for plant-related detection. (A) Electromagnetic enhancement mechanism of plasmonic metal nanoparticles for SERS sensing [87]; (B) charge transfer chemical-enhancement principle of metal semiconductor substrates and structure of fiber-based SERS device [88]; (C) synthetic route of carbon dot lanthanide luminescent probes and Au-nanoparticle-based fluorescent sensing strategy for plant tissue imaging [89].

2.1.1. Fluorescence and FRET Wearable Sensing Principle

Fluorescent wearable patches are commonly functionalized with recognition ligands (e.g., DNA aptamers, antibody fragments, molecularly imprinted polymers [MIPs] [90]) anchored to fluorophores [91,92,93,94]. The fluorescence process generally involves three sequential steps: (i) light absorption (excitation), in which fluorophores absorb incident photons at wavelength λ_ex and undergo electronic transitions from the ground state to higher-energy excited states; (ii) energy loss, in which the excited fluorophores relax to the lowest vibrational level of the first excited singlet state through non-radiative processes such as internal conversion and vibrational relaxation; and (iii) light emission, in which the fluorophores return to the ground state by emitting radiative photons at wavelength λ_em. Radiative and non-radiative transitions compete with each other, and their relative rates determine the fluorescence quantum yield and excited-state lifetime. In receptor-target binding events (e.g., capturing trace stress phytohormones, reactive oxygen species [ROS], or pesticide molecules), the spatial conformation or distance between a donor fluorophore and a proximate acceptor/quencher [95] (e.g., graphene oxide, transition metal dichalcogenides, gold nanoparticles) is modulated, triggering Förster Resonance Energy Transfer (FRET) or Photoinduced Electron Transfer (PET) [96,97,98].

FRET is a non-radiative energy transfer process that occurs between a donor fluorophore and a proximate acceptor/quencher through dipole–dipole coupling, without the emission of a photon. The non-radiative dipole–dipole energy transfer efficiency (E_FRET) is governed by the spatial donor-acceptor separation (R):

$${\mathit{E}}_{\mathit{FRET}} ={\displaystyle \frac{{\mathit{R}}_0^6}{{\mathit{R}}_0^6+{\mathit{R}}^6}}={\displaystyle \frac{1}{1+{\left(\frac{\mathit{R}}{{\mathit{R}}_0}\right)}^6}}$$

(1)

where R₀ is the Förster radius, typically in the range of 2–8 nm, defined as the donor-acceptor distance at which the energy transfer efficiency equals 50%. Because E_FRET decays steeply with increasing R, small changes in donor-acceptor distance caused by target binding can produce large changes in fluorescence readout, which is the basis for FRET-based sensing.

In addition to FRET, fluorescence quenching can also occur through collision or static complexation pathways, which are commonly described by the Stern-Volmer formalism:

$${\displaystyle \frac{{\mathit{I}}_0}{\mathit{I}}}=1+{\mathit{K}}_{\mathit{SV}}\left[\mathit{C}\right]=1+{\mathit{k}}_{\mathit{q}}{\mathit{\tau }}_0\left[\mathit{C}\right]$$

(2)

where I₀ and I are the steady-state fluorescence emission intensities in the absence and presence of the target analyte, respectively; [C] is the analyte concentration; K_SV is the Stern-Volmer quenching constant; kq is the bimolecular quenching rate coefficient; and τ₀ is the unquenched excited-state lifetime of the fluorophore [99]. A key advantage of FRET and quenching-based sensing for wearable applications is that the recognition event is translated into a change in fluorescence intensity or lifetime, which can be measured with miniaturized optical components. For wearable ratiometric fluorophores, dual-emission channels (λ₁, λ₂) calibrate for excitation power fluctuations, ambient illumination drift, and bio-matrix autofluorescence [100,101]. The above FRET sensing principle is shown in Figure 3.

Figure 3. (A) Schematic illustration of the luminescence resonance energy transfer (RET) principle [102]; (B) fabrication process of AMOFs and CNPs, and the smartphone-based fluorescence sensing platform for the detection of CA125/HE4 biomarkers via the FRET “on-off” strategy [103].

2.1.2. SERS Sensing Principle

Surface-enhanced Raman scattering (SERS) is a plasmonic sensing technique that amplifies the Raman scattering signals of molecules adsorbed on or near noble-metal nanostructured surfaces, enabling label-free and fingerprint-like molecular detection with ultrahigh sensitivity. SERS is reviewed here because it provides a powerful optical modality for plant wearable sensors, allowing non-destructive, in situ detection of trace chemical analytes such as pesticide residues, mycotoxins, and pathogen secretions on crop surfaces, which are difficult to measure by conventional electrical sensors. Wearable SERS substrates integrate periodic or quasi-random noble-metal plasmonic nanostructures (e.g., Au/Ag nano-islands, sea-urchin Au@Ag core–shell nanoparticles, 2D MXene/plasmonic superlattices [104]) into breathable, peelable adhesive patches or flexible elastomers [105,106,107]. When excited by monochromatic laser irradiation with incident electric field E_inc, collective oscillations of conduction-band electrons generate Localized Surface Plasmon Resonance (LSPR) [108,109,110,111], inducing extreme spatial confinement of the electromagnetic field within sub-10-nanometer interparticle junctions (“hot spots”) [86,112] (Figure 4).

The total SERS enhancement factor (G_SERS) is governed by the synergistic coupling of the dominant electromagnetic enhancement (G_EM) and the interfacial charge-transfer chemical enhancement (G_CM):

$${\mathit{G}}_{\mathit{SERS}}={\mathit{G}}_{\mathit{EM}}\times {\mathit{G}}_{\mathit{CM}}\approx {\left|\frac{{\mathit{E}}_{\mathit{loc}}\left({\mathit{\omega }}_{\mathit{exc}}\right)}{{\mathit{E}}_{\mathit{inc}}\left({\mathit{\omega }}_{\mathit{exc}}\right)}\right|}^2{\left|\frac{{\mathit{E}}_{\mathit{loc}}\left({\mathit{\omega }}_{\mathit{Raman}}\right)}{{\mathit{E}}_{\mathit{inc}}\left({\mathit{\omega }}_{\mathit{Raman}}\right)}\right|}^2\approx {\left|\frac{{\mathit{E}}_{\mathit{loc}}}{{\mathit{E}}_{\mathit{inc}}}\right|}^4$$

(3)

where E_loc(ω_exc) and E_loc(ω_Raman) denote the locally amplified electric field amplitudes at the excitation and Raman-scattered frequencies, respectively.

Because the Raman scattering cross-section (σ_Raman) is magnified by factors of 10⁶ to 10¹⁴, the collected SERS intensity (ISERS) for a specific vibrational mode at characteristic Raman shift ∆ṽ (cm⁻¹) is expressed as:

I_SERS(∆ṽ) = N_surf∙σ_ads∙I_laser∙G_SERS

(4)

where N_surf is the number of target analyte molecules adsorbed within the electromagnetic hot spots, σ_ads is the modified Raman cross-section upon surface coordination, and I_laser is the incident excitation intensity. Wearable SERS flexible strips enable non-destructive “swab-and-sense” or in situ foliar adhesion, resolving multi-component surface contaminants [113] (e.g., organophosphorus/carbamate pesticides, fungal mycotoxins, and pathogen elicitors) without tedious chromatographic extraction [114,115,116,117].

Figure 4. Fundamental mechanisms and fabrication of plasmonic SERS substrates. (A) Electromagnetic and chemical enhancement mechanisms of SERS: (a) LSPR-induced electromagnetic field enhancement; (b) Energy level diagram of the charge-transfer mechanism. [118]; (B) SERS principle and functional modification of noble-metal nanoprobes [119]; (C) fabrication of Au@MIL-101/PMMA SERS-active substrate [120]; (D)Schematic illustration of enhancement mechanisms: (a) Hot-spot distribution on a metal substrate; (b) Plasmonic coupling between nanoparticles; (c) Electromagnetic enhancement at nanostructure gaps; (d) Charge-transfer enhancement induced by laser excitation [121].

Figure 4. Fundamental mechanisms and fabrication of plasmonic SERS substrates. (A) Electromagnetic and chemical enhancement mechanisms of SERS: (a) LSPR-induced electromagnetic field enhancement; (b) Energy level diagram of the charge-transfer mechanism. [118]; (B) SERS principle and functional modification of noble-metal nanoprobes [119]; (C) fabrication of Au@MIL-101/PMMA SERS-active substrate [120]; (D)Schematic illustration of enhancement mechanisms: (a) Hot-spot distribution on a metal substrate; (b) Plasmonic coupling between nanoparticles; (c) Electromagnetic enhancement at nanostructure gaps; (d) Charge-transfer enhancement induced by laser excitation [121].

2.2. Fundamental Principles of Electrical Wearable Sensors

Electrical wearable sensors convert biochemical redox events, ion activities, and mechanical strain into quantifiable electrical metrics (current, potential, resistance, capacitance) using conductive nanomaterial networks (e.g., laser-induced graphene [LIG] [122], carbon nanotubes [CNTs], MXenes, conductive polymers like PEDOT:PSS, and liquid metals) [123,124,125]. Their intrinsic solid-state operation allows seamless integration with miniaturized signal conditioning chips, Bluetooth/NFC/LoRa wireless telemetry, and energy harvesting components [126] (Figure 5).

Figure 5. Schematic overview of electrical wearable sensors. (A) Structure and piezoresistive sensing performance of ANFs/CNTs composite wearable sensor [127]; (B) GO-based sensing workflow combined with machine learning analysis for plant signal recognition [128]; (C) fabrication of modified microelectrodes for in situ plant detection [129]; (D) hardware circuits for resistive, capacitive and nanogenerator sensors [130].

Figure 5. Schematic overview of electrical wearable sensors. (A) Structure and piezoresistive sensing performance of ANFs/CNTs composite wearable sensor [127]; (B) GO-based sensing workflow combined with machine learning analysis for plant signal recognition [128]; (C) fabrication of modified microelectrodes for in situ plant detection [129]; (D) hardware circuits for resistive, capacitive and nanogenerator sensors [130].

2.2.1. Electrochemical Wearable Sensing Principle

Wearable electrochemical platforms typically implement a planar or microneedle-based three-electrode architecture [131,132,133]: a functionalized Working Electrode (WE), a Counter Electrode (CE), and a Reference Electrode (RE, such as Ag/AgCl pseudo-reference) [134,,135]. Depending on the transduction mode, these are classified into amperometric/voltammetric, potentiometric, and impedimetric sensors:

Amperometric Sensing: Target analytes (e.g., electroactive phytohormones like salicylic acid, ascorbic acid, or enzymatic hydrogen peroxide byproduct from glucose/oxidase cascades) undergo diffusion-controlled Faradaic redox reactions at the WE under an applied overpotential η [136,137,138]. For an ideal planar electrode with infinite size, where diffusion is strictly one-dimensional, the time-dependent Faradaic current transient I(t) is governed by the classical Cottrell equation:

$$\mathit{I}\left(\mathit{t}\right)={\displaystyle \frac{\mathit{nFA}{\mathit{D}}^{1/2}{\mathit{C}}_{\mathit{bulk}}}{{\mathit{\pi }}^{1/2}{\mathit{t}}^{1/2}}}$$

(5)

Under convective microfluidic or steady-state radial mass transport conditions (e.g., microelectrode/microneedle arrays), the steady-state Faradaic current (I_SS) correlates linearly with the bulk analyte concentration:

$${\mathit{I}}_{\mathit{SS}}={\mathit{nFAk}}_{\mathit{m}}{\mathit{C}}_{\mathit{bulk}}={\displaystyle \frac{\mathit{nFAD}}{{\Delta }_{\mathit{N}}}}{\mathit{C}}_{\mathit{bulk}}$$

(6)

where n is the number of transferred electrons per molecule, F is the Faraday constant (96,485 C/mol), A is the electrochemically active surface area (cm²), D is the molecular diffusion coefficient (cm²·s⁻¹), k_m is the mass-transport transfer coefficient (cm·s⁻¹), and δ_N is the Nernst diffusion layer thickness (cm). It should be noted that Equations (5) and (6) are strictly valid for one-dimensional diffusion at planar electrodes with infinite size. In practical wearable electrochemical sensors, electrode geometries such as microelectrodes, microneedle arrays, interdigitated array (IDA) electrodes, and circular electrodes are frequently used, where the diffusion field becomes non-planar and may approach hemispherical or cylindrical symmetry. Under such conditions, the amperometric response deviates from the Cottrell behavior and can reach a true steady state dominated by radial diffusion, with the steady-state current given by:

Iss,radial = 4nFDC_bulkr_d

(7)

for a microdisk electrode of radius r_d, or by geometry-dependent expressions for IDA and circular electrodes. In these cases, the current is no longer proportional to t^(−1/2) and is less sensitive to convection, which is advantageous for continuous in situ monitoring on plant surfaces. Therefore, the electrode geometry and diffusion regime should be carefully considered when applying Equations (5) and (6) to practical wearable electrochemical sensors.

Potentiometric Ion-Selective Sensing: Solid-contact ion-selective electrodes (SC-ISEs) quantify specific ionic activities (e.g., K⁺, Na⁺, NO₃⁻, Ca²⁺, H⁺ for pH) in apoplastic fluids or xylem sap under zero net Faradaic current (I = 0) [139,140,141]. The electromotive force (E_cell) obeys the extended Nicolsky-Eisenman equation:

$$E_{\mathit{cell}}={\mathit{E}}^0+{\displaystyle \frac{2.303{\mathit{R}}_{\mathit{g}}\mathit{T}}{{\mathit{z}}_{\mathit{i}}F}}{\log }_{10}\left[a_i+{\sum }_j{K}_{i,j}^{\mathit{pot}}{\left(a_j\right)}^{{\mathit{z}}_{\mathit{i}}/{\mathit{z}}_{\mathit{j}}}\right]$$

(8)

where E⁰ is the standard cell potential, R_g is the universal gas constant (8.314 J·mol⁻¹K⁻¹), T is absolute temperature (K), Z_i and Z_j are charges of the primary ion i and interfering ion j, a_i and a_j are their respective chemical activities, and $K_{i,j}^{\mathit{pot}}$ is the potentiometric selectivity coefficient.

Impedimetric Sensing and Electrochemical Impedance Spectroscopy (EIS): Wearable impedimetric sensors interrogate interfacial bio-recognition dynamics, ion permeability [142,143,144], and vegetal tissue status by applying a small-amplitude sinusoidal perturbation potential E(t) = E₀sin(ωt) (typically 5~10 mV) across a broad angular frequency spectrum ω = 2πf, recording the resulting phase-shifted sinusoidal current I(t) = I₀sin(ωt + θ) [145,146]. The complex impedance Z(ω) is expressed in Cartesian and polar forms as:

$$\mathit{Z}\left(\mathit{\omega }\right)={\displaystyle \frac{\mathit{E}\left(\mathit{t}\right)}{\mathit{I}\left(\mathit{t}\right)}}=Z^{\prime }\left(\omega \right)+\mathit{j}{\mathit{Z}}^{″}\left(\mathit{\omega }\right)=\left|Z\right|\exp (j\mathrm{\theta })$$

(9)

where Z′ is the real (resistive/in-phase) component, Z″ is the imaginary (reactive/out-of-phase) component, $j=\sqrt{-1}$ is the imaginary unit, $\left|Z\right|=\sqrt{(Z^{\prime }{)}^2+(Z^{″}{)}^2}$ is the impedance magnitude, and θ = arcttan(Z″/Z′) is the phase angle.

At the bio-electrode interface, interfacial charge transfer kinetics are quantitatively modeled using the classic Randles equivalent electrical circuit, which incorporates the uncompensated electrolyte/solution resistance (R_s), the interfacial double-layer capacitance (C_dl), or constant phase element (CPE), the charge-transfer resistance (R_ct), and the Warburg diffusion impedance (Z_W):

$${\mathit{Z}}_{\mathit{Randles}}\left(\mathit{\omega }\right)={\mathit{R}}_{\mathit{s}}+{\displaystyle \frac{1}{\mathit{j\omega }{\mathit{C}}_{\mathit{dl}}+\frac{1}{{\mathit{R}}_{\mathit{ct}}+{\mathit{Z}}_{\mathit{W}}\left(\mathit{\omega }\right)}}}$$

(10)

where the semi-infinite linear diffusion Warburg impedance is governed by:

$${\mathit{Z}}_{\mathit{W}}\left(\mathit{\omega }\right)={\mathit{\sigma }}_{\mathit{W}}{\mathit{\omega }}^{-{\displaystyle \frac{1}{2}}}(1-\mathit{j})$$

(11)

where σ_W is the Warburg coefficient related to ionic diffusion coefficients. It should be noted that Equations (10) and (11) are strictly valid for an ideal one-dimensional semi-infinite diffusion geometry in an infinitely long electrochemical cell. In practical plant-wearable sensors, this idealized model is often insufficient. First, the electrode-tissue interface is finite in size, and diffusion may become non-planar (e.g., radial or hemispherical) when microelectrodes, microneedle arrays, or interdigitated electrodes are used. Second, the plant tissue itself contributes additional impedance components, including cell membrane capacitance, intracellular and extracellular fluid resistance, and apoplastic/cytoplasmic ion transport pathways, which are commonly described by tissue-level Cole–Cole or distributed-element models rather than a simple Warburg element. Third, the presence of cuticular wax, biofouling layers, and non-uniform contact between the sensor and the plant surface introduces additional interfacial impedance and heterogeneity. Therefore, in practical plant-wearable impedance sensing, the Randles circuit and the Warburg expression should be regarded as simplified limiting cases, and the equivalent circuit should be extended with geometry-dependent diffusion elements, tissue-level impedance components, and interfacial heterogeneity terms to better describe the actual sensor–plant interface.

In label-free affinity biosensing (e.g., detecting phytopathogen spores, specific viral proteins, or pesticide binding), the formation of target-probe immunocomplexes or aptamer-target folding at the WE interface introduces steric hindrance and electrostatic repulsion against redox probes (e.g., [Fe(CN)₆]^3−/4−), resulting in a quantifiable increase in R_ct (manifested as an expanding semicircle diameter in the high-frequency region of the Nyquist plot) [147].

Concurrently, in in vivo physiological monitoring, wearable microneedles or surface-grafted multielectrodes employ non-Faradaic EIS (utilizing tissue-level Cole-Cole models) to map intracellular fluid resistance (R_i), extracellular fluid resistance (R_e), and cell membrane capacitance (C_m) [148,149]. This enables continuous tracking of plant tissue water status, drought-induced cellular plasmolysis, and early membrane permeability damage induced by cold, salinity, or fungal invasion.

Microneedle-integrated electrochemical arrays minimally penetrate the hydrophobic cuticular wax barrier without eliciting systemic wound defense cascades, achieving real-time, multiplexed biofluid profiling in vivo [150] (Figure 6).

Figure 6. Representative electrochemical sensing platforms. (A) Electrochemical recognition and potentiometric ratiometric signal strategies [151]; (B) catalytically modified electrode for voltammetric detection [152]; (C) microneedle-based three-electrode sensor for in situ plant real-time monitoring [153]; (D) Electrochemical characterization and sensing performance: (a,b) CV curves at different DA concentrations using traditional (a) and integrated (b) electrode systems; (c) Calibration curves showing the linear relationship between current change and DA concentration; (d) Selectivity evaluation of the sensor against various interfering substances. [154].

2.2.2. Piezoresistive Wearable Sensing Principle

Piezoresistive sensors monitor mechanical strains induced by stem expansion/shrinkage, foliar turgor oscillation, and robotic grasping forces [155,156,157]. The sensitive layer comprises conductive nanofillers (e.g., graphene, CNTs, MXene nanosheets [158], metal nanowires) dispersed within or coated onto flexible elastomeric matrices (e.g., PDMS, Ecoflex, TPU, natural rubber) [159,160,161].

The primary figure of merit is the Gauge Factor (GF), defined as the normalized relative resistance change per unit mechanical strain (ε):

$$\mathit{GF}={\displaystyle \frac{\Delta R/R_0}{\epsilon }}={\displaystyle \frac{(R-R_0)/R_0}{\Delta L/L_0}}$$

(12)

where R₀ and R represent the electrical resistances in the relaxed and strained states, and L₀ and ΔL denote the initial length and elongation of the sensor, respectively.

Mechanistically, the total electrical conductivity (σ_cond) near the percolation threshold (ϕ_c) is described by classical percolation theory and Simmons’ quantum tunneling model:

$${\mathit{\sigma }}_{\mathit{cond}}\propto {\left(\mathit{\varphi }-{\mathit{\varphi }}_{\mathit{c}}\right)}^t \mathit{for} \mathit{\varphi }>{\mathit{\varphi }}_{\mathit{c}}$$

(13)

$${\mathit{R}}_{\mathit{tunnel} }={\displaystyle \frac{{\mathit{h}}^2\mathit{d}}{{\mathit{A}}_{\mathit{c}}{\mathit{e}}^2\sqrt{2{\mathit{m}}_{\mathit{e}}\overline{\mathit{\phi }}}}}\exp ({\displaystyle \frac{4\mathit{\pi d}}{\mathit{h}}}\sqrt{2{\mathit{m}}_{\mathit{e}}\overline{\mathit{\phi }}})$$

(14)

where ϕ is the conductive filler volume fraction, t is the universal transport exponent, d is the interparticle tunneling distance, A_c is the effective contact area, h is Planck’s constant, e is the elementary electron charge, m_e is the effective electron mass, and $\overline{\phi }$ is the mean potential barrier height between adjacent conductive islands.

Under applied mechanical stress σ_stress within the linear elastic regime (governed by Young’s modulus E_Y, such that ε = σ_stress/E_Y), the resistance modulation is formulated as:

$${\displaystyle \frac{\Delta \mathit{R}}{{\mathit{R}}_0}}=\mathit{GF}\times {\displaystyle \frac{{\mathit{\sigma }}_{\mathit{stress}}}{{\mathit{E}}_{\mathit{Y}}}}$$

(15)

Wearable piezoresistive bands attached to crop stalks yield micro-displacement resolution (<1 μm), directly mapping diurnal stem hydration dynamics and drought stress cycles [162,163,,164].

2.2.3. Capacitive Wearable Sensing Principle

Capacitive wearable sensors monitor variations in dielectric permittivity, electrode spacing, or interfacial double-layer capacitance under physiological swelling, foliar transpiration, or physical contact [165,166,167]. A parallel-plate capacitive architecture exhibits a baseline capacitance (C₀):

$${\mathit{C}}_0={\mathit{\epsilon }}_0{\mathit{\epsilon }}_{\mathit{r}}{\displaystyle \frac{{\mathit{A}}_{\mathit{eff}}}{\mathit{d}}}$$

(16)

where ε₀ is the vacuum permittivity (8.854 × 10⁻¹² F·m⁻¹), ε_r is the relative dielectric constant of the dielectric interlayer, A_eff is the effective overlapping plate area, and d is the dielectric separation thickness.

Under foliar transpiration or humidity fluctuations, water vapor molecule adsorption/desorption dynamically modulates the effective relative dielectric permittivity, governed by the Maxwell–Garnett effective medium approximation:

$${\displaystyle \frac{{\mathit{\epsilon }}_{\mathit{r},\mathit{eff}}-{\mathit{\epsilon }}_{\mathit{poly}}}{{\mathit{\epsilon }}_{\mathit{r},\mathit{eff}}+2{\mathit{\epsilon }}_{\mathit{poly}}}}={\mathit{\theta }}_{\mathit{w}}\left({\displaystyle \frac{{\mathit{\epsilon }}_{\mathit{water}}-{\mathit{\epsilon }}_{\mathit{poly}}}{{\mathit{\epsilon }}_{\mathit{water}}+2{\mathit{\epsilon }}_{\mathit{poly}}}}\right)$$

(17)

where ε_water ≈ 80, ε_poly ≈ 2~4 represents the dry polymer substrate permittivity, and θ_w is the adsorbed volumetric water fraction.

When configured with ionic conductors (e.g., ionic hydrogels, ionogels), an Electrical Double Layer (EDL) forms at the electrode-electrolyte interface with an ultra-high specific capacitance ($C_{\mathit{EDL}}={\epsilon }_0{\epsilon }_{\mathrm{r}}{\displaystyle \frac{A_{\mathrm{eff}}}{{\mathrm{\lambda }}_D}}$), where λ_D is the sub-nanometer Debye screening length). Under normal compressive stress, the dynamic increase in microscopic contact area induces substantial capacitance changes (ΔC/C₀) [168,169,170], providing ultra-high sensitivity and minimal hysteresis for soft robotic fruit picking and leaf turgor tracking (Figure 7).

Figure 7. Flexible capacitive wearable sensor platforms. (A) Chemical components, nanofiber morphology and pressure-responsive EDL capacitance mechanism of ion-gel-based sensor [171]; (B) assembly process, mechanical simulation, performance characterization and micro-structure of fabric-sponge capacitive sensor [172].

2.3. Comprehensive Comparison and Strategic Modality Selection

The selection of optical versus electrical wearable sensing modalities involves balancing analytical sensitivity, selectivity, physical footprint, power budget, and field deployment logistics. Moreover, Table 2 further distinguishes the suitable field-application scenarios and practical advantages of optical and electrical wearable sensors to guide modality selection for specific agricultural tasks.

Table 2. Comprehensive comparison of wearable optical and electrical sensing modalities in precision agriculture.

Comparison Items	Optical Wearable Sensors	Electrical Wearable Sensors
Core Advantages	Capable of remote and non-contact measurement; high sensitivity for trace analytes; label-free molecular detection; suitable for early molecular-level warning	Mature and standardized electrical readout; compatible with existing wireless transmission and signal-conditioning circuits; suitable for continuous, long-term, in situ monitoring
Main Limitations	Light excitation typically requires relatively higher power than purely electronic readout; susceptible to ambient light, autofluorescence and matrix interference	Generally require direct contact with plant tissue, which increases manual deployment effort and exposes the interface to biofouling; susceptible to pH, impurity-ion and plant-matrix interference
Subtype Classification	Luminescence sensors (fluorescence, bioluminescence, chemiluminescence, etc.); Raman scattering sensors (SERS)	Electrochemical sensors (amperometric/voltammetric, potentiometric, impedimetric); piezoresistive sensors; capacitive sensors
Suitable field-application scenarios	Early diagnosis of crop diseases and abiotic stress; rapid safety screening of postharvest agricultural products (pesticide residues, mycotoxins, pathogens); non-destructive ripeness and quality assessment	Long-term monitoring of soil and plant sap nutrients and water status; crop growth strain and transpiration; intelligent harvesting tactile/pressure feedback; grain moisture and mildew risk monitoring
Practical advantages	Remote, non-contact and label-free molecular detection; suitable for trace biomarkers and chemical contaminants	Low cost, simple structure, easy integration with mature electrical readout and wireless modules; suitable for continuous, long-term, in situ field monitoring and robotic control

Beyond the comparison summarized in Table 1, substrate adaptability is another critical but often overlooked factor determining the practical deployment of plant wearable sensors across diverse crop species. Grain crops (e.g., rice, wheat, maize) generally possess narrow, slender, and highly curved leaves with dense epicuticular wax crystals, weak mechanical support, and rapid leaf elongation and senescence, requiring ultra-thin, low-modulus, and highly conformal substrates that can adhere stably without restricting leaf movement or photosynthesis [173]. In contrast, fruit-vegetable plants (e.g., tomato, cucumber, strawberry, peach) exhibit broader leaves, thicker stems, expanding fruits, and pubescent or waxy organ surfaces, where sensor substrates must simultaneously accommodate large curvature changes, surface growth-induced strain, and fruit expansion without delamination or signal drift. Moreover, the mechanical mismatch between rigid conductive nanomaterials and soft plant tissues differs markedly among species: stem-based piezoresistive bands require high stretchability and fatigue resistance for woody or succulent stems, whereas leaf-attached fluorescent or SERS patches demand optical transparency, breathability, and minimal mass loading to avoid altering leaf microclimate and transpiration. Therefore, a single universal flexible substrate is unlikely to satisfy all crop types. Future substrate design should shift toward species-adaptable strategies, including tunable modulus and thickness, bio-inspired adhesive microstructures, degradable or breathable matrices, and crop-specific conformal layouts, so that wearable sensors can achieve stable long-term adhesion and reliable signal transduction across both grain crops and fruit-vegetable plants [174].

3. Applications for Crop Growth Monitoring

Conformal plant wearable sensors enable continuous, non-invasive, and in situ tracking of vital physiological indicators, systemic signaling dynamics, and external environmental stresses throughout the crop lifecycle [175,176]. By translating invisible cellular and biomechanical responses into quantifiable optical and electrical outputs, wearable sensors eliminate the latency and destructive sampling inherent to conventional offline assays (Figure 8).

Figure 8. Practical applications of plant wearable sensors for crop physiological monitoring. (A) The overall workflow of on-plant wearable sensing, real-time data acquisition and intelligent classification analysis [177]; (B) time-dependent resistance response signals after stem and leaf cutting damage [177]; (C) performance comparison between a wearable PlantRing sensor and a gravimetric system for drought stress monitoring [178]; (D) sensing mechanism for rapid detection of organophosphorus pesticide residues [179].

3.1. Detection of Crop Biochemical and Physiological Biomarkers

Reactive oxygen species (ROS) and endogenous phytohormones serve as primary biochemical sentinels that orchestrate plant systemic defense signaling long before macroscopic phenotypic symptoms emerge [180,181,182]. Under acute abiotic stresses such as drought shock, thermal extremes, and heavy metal exposure, or during early phytopathogenic invasion [183,184,185], plants undergo rapid oxidative bursts characterized by trace accumulation of hydrogen peroxide (H₂O₂) and superoxide anions (·O₂⁻) [186,187]. Conformal fluorescent wearable patches functionalized with target-specific DNA aptamers or catalytic nanozymes adhere directly to the foliar epidermis [99,188,189,190]. The selective binding of ROS dynamically changes the donor-to-acceptor spatial distance, thereby triggering FRET, yielding instantaneous emission quenching or spectral shifts proportional to picomolar analyte concentrations [191,192].

In parallel, dynamic phytohormone fluxes regulate essential developmental and defense pathways, where abscisic acid (ABA) governs stomatal aperture and foliar transpiration under water deficit [153,193], while salicylic acid (SA) and jasmonic acid (JA) coordinate systemic acquired resistance against biotrophic and necrotrophic pathogens. To track these internal signaling molecules without inducing defensive wounding cascades [194,195], flexible polymeric microneedles minimally penetrate the hydrophobic cuticular wax layer to interface directly with apoplastic sap. Working electrodes modified with electrocatalytic nanocomposites [196,197], such as MXenes, carbon nanotubes, and metallic nanoparticles, perform Faradaic amperometric and differential pulse voltammetric measurements, resolving sub-micromolar fluctuations of ABA and SA in real time to provide an actionable diagnostic window for timely agronomic interventions [198,199,200].

3.2. Dynamic Monitoring of Abiotic and Biotic Stresses

Wearable sensors realize synchronous monitoring of two major crop stress types. Capacitive leaf sensors reflect crop transpiration and water shortage via real-time humidity data [201,202,203]. The working mechanism is based on parallel-plate capacitor theory: leaf surface water vapor changes the relative dielectric constant of the intermediate dielectric layer of the capacitor, and capacitance variation output corresponds to leaf water content, realizing continuous humidity response [204]. Piezoresistive stem bands record daily stem shrinkage and expansion rhythms to evaluate long-term drought stress [205,,206,207]. The core sensing principle is the piezoresistive effect: tiny volume deformation of stems changes the contact area of internal conductive fillers in composite sensitive materials, leading to regular resistance change, and deformation degree can be calculated by the material gauge factor. SERS optical wearable sensors identify trace pathogen secretions on leaf surfaces to achieve early diagnosis of fungal and bacterial blight [208,209]. The detection mechanism originates from local surface plasmon resonance (LSPR): laser excitation generates strong electromagnetic hot spots on noble metal nanostructures, which amplifies molecular Raman scattering signals of secretions, and unique Raman shift peaks correspond to different pathogenic metabolites [210,211,212]. For crop phenotyping, long-term attached strain sensors record stem thickening and fruit expansion rates, matching field meteorological data to establish dynamic crop growth curves, supporting precise variable fertilization and irrigation decision-making [50,213,214,215,216].

3.3. High-Throughput Phenotyping and Precision Agronomic Feedback

Real-time physiological monitoring by wearable sensors effectively reduces global crop yield losses caused by various stresses [217,218]. All the above wearable sensing systems rely on stable optical or electrical signal transduction mechanisms to convert invisible plant physiological and morphological changes into readable quantitative signals, eliminating the lag and destructive defects of traditional offline detection methods. Precise water and fertilizer supply guided by in situ physiological signals avoids blind excessive input, saving water resources and alleviating chemical fertilizer non-point source pollution [219,220,221]. In addition, wearable sensors realize high-throughput continuous phenotyping of single plants, precisely distinguishing stress resistance differences in different germplasm resources and shortening the screening cycle of stress-resistant crop varieties. The unified signal conversion principle enables long-term, continuous, single-plant parallel monitoring, which cannot be realized by discrete laboratory detection means, laying a technical foundation for large-scale crop germplasm screening and field stress dynamic tracking [222,223,224]. It should be noted that the wearable sensing applications reviewed here are not intended to be universally standardized across all agricultural regions. Instead, they should be adapted to local crop species, climatic conditions, soil types, farming practices, and socioeconomic contexts. For example, the same optical or electrical sensing platform may need different calibration models, adhesion strategies, and data interpretation pipelines when deployed on rice paddies in humid subtropical regions versus wheat fields in semi-arid temperate regions. Local wisdom, such as traditional knowledge of crop phenology, irrigation timing, and pest outbreak patterns, can provide valuable priors for sensor deployment, signal interpretation, and decision-making, and should be integrated with standardized sensor hardware and algorithms to improve practical applicability. Beyond regional adaptation, the economic value of wearable sensor deployment is also a key consideration for practical adoption. From an economic perspective, the adoption of wearable sensors can generate measurable benefits through higher and more stable yields as well as reduced input and postharvest losses. By enabling early stress detection and precise irrigation and fertilization, wearable sensors can help maintain crop yield under adverse conditions and reduce input costs associated with excessive water, fertilizer, and pesticide application. For example, in-field physiological monitoring can support variable-rate irrigation and site-specific fertilization, which have been reported to improve water use efficiency and nutrient use efficiency while maintaining or increasing yields. In addition, early warning of disease and abiotic stress allows timely and targeted intervention, reducing yield losses and postharvest waste. These economic gains are particularly important for smallholder farms, where even modest improvements in yield stability and input efficiency can translate into significant income increases. However, the economic value of wearable sensors also depends on device cost, calibration and maintenance requirements, and the availability of data interpretation services, which should be considered in future cost–benefit analyses.

4. Applications in Harvesting and Agricultural Product Quality Safety

Benefiting from the development of wearable nanosensing technology, compact and highly sensitive sensors satisfy the on-site real-time monitoring demands during crop harvesting and subsequent quality safety management [225,226]. Pressure sensing components equipped on robotic picking manipulators supply dynamic gripping force feedback, which effectively prevents mechanical damage to delicate fruits in automated harvesting [227,228,229]. Moreover, these flexible sensing tools enable non-destructive identification of multiple quality traits including ripeness, firmness and freshness indicators without sample destruction [230]. They are also capable of rapidly sensing and analyzing common safety risks such as pesticide residues, microbial contamination and toxic pollutants [231,232,233]. This section accordingly reviews three primary application directions, covering pressure perception for intelligent fruit-picking machinery, non-destructive agro-product quality inspection and on-site sensing-based agricultural product safety analysis. Similarly, in harvesting and postharvest quality/safety monitoring, sensor deployment should be adapted to local agricultural practices and market requirements. For instance, pesticide residue screening thresholds, ripeness criteria, and storage conditions may vary among regions and crop varieties. Therefore, standardized sensing principles and device architecture should be combined with region-specific calibration, threshold setting, and decision rules, rather than being applied uniformly across all agricultural systems. In harvesting and postharvest operations, wearable sensors can also deliver direct economic benefits by reducing mechanical damage during picking, improving sorting and grading accuracy, and enabling early detection of quality deterioration. These improvements can increase the proportion of marketable produce, extend storage life, and reduce postharvest losses, thereby raising the overall economic return for growers and supply-chain stakeholders.

4.1. Tactile Pressure Sensing for Lossless Robotic Harvesting

Piezoresistive flexible sensing films are integrated on clamping terminals of automatic picking robots and flexible picking gloves [234,235,236,237]. During picking operations, the sensor transmits real-time extrusion pressure data to the control terminal [238,239,240]. When the pressure exceeds the peel damage threshold, the manipulator automatically reduces clamping force to realize lossless picking of fragile fruits such as strawberries and peaches [241,242,,243]. For manual picking scenarios, wearable pressure gloves remind farmers of excessive holding force through vibration feedback, lowering the mechanical damage rate and postharvest rot loss of fruits [244,245,246,247]. The structure, working mechanism, performance characterization and practical application demonstrations of the MXene/leather piezoresistive pressure sensor are displayed in Figure 9.

Figure 9. MXene/leather flexible piezoresistive pressure sensor. (A) Device structure, pressure-responsive working mechanism, gradient-pressure dynamic response, optical photographs, sensitivity calibration, step-response, cyclic stability and pressure loading setup tests of the sensor [248]; (B) electrical output signals obtained from air pressure detection tests [249]; (C) practical application test for soft robotic grasping [250].

4.2. Non-Destructive Quality Phenotyping and Ripeness Assessment

Optical wearable sensing stickers attached to fresh fruits and vegetables complete full-process non-destructive quality detection after harvest. Fluorescence sensors identify fruit ripeness grades according to chlorophyll and sugar fluorescence signal changes, supporting classified picking and storage [251]. Flexible SERS patches rapidly detect soluble solids and titratable acids on fruit surfaces without peeling to evaluate flavor quality [252,253,254]. For grain storage, portable wearable sensors monitor real-time grain moisture content and give early warning of mildew risks, fully compatible with high-throughput assembly line screening without destroying agricultural products [255,256,257]. Typical wearable sensing devices for agricultural product quality inspection and plant physiological monitoring are shown in Figure 10.

Figure 10. Schematic diagram of various wearable sensing systems for agricultural applications. (A) Preparation process of SWNT/PVA fiber humidity sensor [258]; (B) microneedle sensor for in situ plant leaf pH real-time monitoring [259]; (C) general framework of three classic electrochemical sensing detection modes [260]; (D) fabrication and working principle of triboelectric sensor for fruit freshness evaluation [261].

4.3. On-Site Trace Detection of Agrochemical Residues and Contaminants

The core detection targets of agricultural product safety include pesticide residues, heavy metal pollution and toxic mycotoxins [262,263,264]. Flexible SERS sensing stickers attached to the fruit epidermis capture trace organophosphorus and carbamate pesticide residues, completing qualitative and quantitative analysis via characteristic Raman peaks [265]. Aptamer-modified fluorescence wearable sensors detect aflatoxin and ochratoxin in grain extracts with picomolar detection limits, meeting strict food safety standards [266,267,268,269]. Different from large laboratory chromatographic instruments, portable wearable devices support on-site rapid screening in orchards, farmers’ markets and processing plants, blocking unqualified agricultural products at the source [270,271,272]. A series of optical and electrochemical sensing platforms for agricultural safety detection are displayed in Figure 11.

Figure 11. Various sensing platforms for agricultural product safety detection. (A) Fluorescent sensing film for pesticide residue visual monitoring [273]; (B) preparation and DPV response of two MXene-based electrochemical pesticide sensors [260]; (C) working flowchart of fiber optic probes for pathogenic bacteria identification in agro-products [274]; (D) N-CQDs fluorescent system for carbamate pesticide quantitative detection [275].

5. Signal Processing Technologies for Sensing Data

Raw data acquired by wearable nanosensors frequently suffers from ambient noise, baseline drift and miscellaneous interference under complicated field test environments, so the primitive electrical responses are incapable of supporting agricultural evaluation directly. Effective signal processing acts as an essential intermediate procedure to translate sensor outputs into actionable agricultural information. The initial signal preprocessing removes unwanted interference and improves the quality of unprocessed sensing data. Subsequent feature extraction selects valuable information, while data calibration compensates for sensor deviation triggered by temperature, humidity and complex agricultural sample matrices. After that, machine learning-based intelligent analysis achieves qualitative recognition and quantitative estimation for crop growth status, fruit quality and food safety risks [276,277,278]. This section illustrates three sequential processing phases, which cover the pretreatment of original sensing signals, feature extraction and data calibration, as well as intelligent interpretation for multi-source agricultural sensing signals.

5.1. Preprocessing of Raw Sensing Signals

Original signals collected by optical and electrical wearable sensors contain multiple forms of interference noise originating from complex farm environments, including slow temperature drift, dielectric disturbance from rainwater and high-frequency mechanical clutter from field machinery operation, and signal preprocessing serves as an essential preliminary analytical step covering denoising, drift correction and outlier elimination [279,280,281]. Denoising relies on moving average filtering and wavelet threshold filtering to erase random high-frequency noise, with the time-series moving average formula defined as:

$$\overline{\mathrm{x}}(\mathrm{t})={\displaystyle \frac{1}{2\mathrm{N}+1}}{\displaystyle {\sum }_{\mathrm{i}=-\mathrm{N}}^{\mathrm{N}}}\mathrm{x}(\mathrm{t}+\mathrm{i})$$

(18)

where 2N + 1 stands for an adjustable filter window length tuned according to signal sampling frequency, and drift correction builds temperature compensation models using real-time field temperature readings to counteract slow signal decay from long-term sensitive layer aging, while the three-sigma rule removes anomalous signal points caused by temporary poor contact between sensor and crop surface to avoid distortion of subsequent quantitative analysis outcomes. Multi-step preprocessing retains stable baselines and complete effective characteristic response signals to establish clean, reliable datasets for follow-up feature extraction and calibration modeling [282].

5.2. Feature Extraction and Matrix Calibration

Feature extraction extracts effective characteristic parameters closely related to detection targets from denoised continuous signals. For electrical sensing data, steady-state current, resistance variation amplitude and capacitance change rate are extracted as core features. For optical signals, fluorescence peak intensity and Raman shift values are selected. Time-domain features such as signal response duration are extracted from long-term crop monitoring data to reflect dynamic physiological variation processes. Matrix matching calibration and standard addition methods are applied to eliminate interference from plant pigments and fruit organic acids, establishing linear fitting curves between signal features and standard analyte concentrations to improve quantitative detection accuracy in complex samples [283,284,285].

5.3. Intelligent Analysis and Edge-Cloud Symbiosis

After feature extraction and calibration, machine learning and deep learning algorithms are introduced for intelligent identification and prediction [286,287,288]. Support vector machine and random forest models classify crop disease types and fruit ripeness grades based on multi-dimensional sensing features. Lightweight CNN and LSTM networks are deployed on wearable edge chips to realize local real-time analysis of time-series signals [289], automatically identifying drought stress and grain mildew risks without full raw data uploading to reduce transmission delay and power consumption [290,291,292]. Cloud big data platforms fuse multi-point wearable data, soil and meteorological information to construct crop growth prediction models and output optimized irrigation and pesticide application schemes for precision agriculture management [286,293,294,295]. The mainstream intelligent signal analysis algorithms and corresponding test results are shown in Figure 12.

Figure 12. Schematic of intelligent signal analysis algorithms for agricultural sensing data. (A) Time series prediction frameworks based on HMM and LSTM: (a) HMM state transition diagram; (b) Typical current signal time series; (c) LSTM-based neural network architecture; (d) Comparison of raw, ground truth, and deep-channel output signals. [296]; (B) Random forest modeling workflow [297]; (C) Chemometric analysis including PCA dimensionality reduction, clustering and discriminant classification: (a–c) PCA score plots with different numbers of principal components (n_components = 1, 2, 3); (d) Schematic of clustering process; (e) Illustration of maximum interval and optimal hyperplane; (f) Schematic of distance metrics (components = 2); (g) Multi-dimensional SERS data acquisition and chemometric analysis (PCA, LDA) for tea recognition; (h) Confusion matrix showing classification sensitivity and specificity; (i–j) Schematic of colorimetric sensor array and its corresponding PCA score plot; (k) Decision tree classification model workflow [298].

6. Conclusions and Outlook

6.1. Conclusions

This review systematically sorts out two mainstream categories of agricultural wearable sensors: optical and electrical, and elaborates their respective optical and electrical signal transduction mechanisms. Optical wearable sensors rely on fluorescence and SERS effects to achieve ultra-sensitive trace detection of biological metabolites and pollutants. Electrical wearable sensors feature simple structures and low costs, suitable for long-term continuous monitoring of crop growth and environmental indices, with distinct applicable field scenarios for both types.

Wearable sensing technology covers the whole agricultural industrial chain. During crop growth periods, it realizes early stress warning and high-throughput phenotyping by detecting physiological markers. In post-harvest picking and processing links, lossless intelligent clamping control and rapid quality and safety screening of agricultural products can be completed. A complete signal processing system including noise reduction, feature extraction and edge intelligent analysis effectively suppresses complex farmland environmental interference and improves the practicability of field deployment. At present, the popularization of agricultural wearable sensors is still restricted by single sensing mode, poor long-term environmental stability and high preparation cost of nanomodified sensitive materials.

6.2. Outlook

Future research should prioritize multimodal integrated wearable devices that combine optical fluorescence/SERS and electrochemical/piezoresistive sensitive layers on a single flexible substrate to achieve synchronous multi-parameter testing of crop physiology, growth strain and field environmental conditions and break the single-signal limitation of current single-function hardware, alongside bionic self-powered wearable sensors integrated with piezoelectric and triboelectric nanogenerator energy harvesting units that capture wind vibration and stem expansion mechanical energy to power sensing modules and eliminate dependence on disposable lithium batteries for over 30 days of unattended field monitoring. Biomimetic nanosensors refer to sensing devices that mimic biological sensory mechanisms (e.g., plant olfactory or tactile systems) by integrating bio-recognition elements (such as odorant-binding proteins, aptamers, or molecularly imprinted polymers) with nanomaterial transducers, enabling highly selective and sensitive detection of trace analytes [299,300]. AI-coupled agricultural monitoring refers to the integration of artificial intelligence algorithms (e.g., machine learning, deep learning, edge computing) with wearable sensor data streams to enable real-time signal processing, pattern recognition, and predictive decision-making for crop management [301,302]. Researchers should also pay special attention to the anti-biofouling bottleneck of plant-attached sensors under harsh farm conditions. Current surface modification strategies still face notable limitations: most superhydrophobic or anti-fouling coatings suffer from insufficient long-term stability under cyclic UV radiation, rain washing, agrochemical erosion, and mechanical deformation, which can degrade low-surface-energy layers or destroy micro/nanostructures; moreover, farmland biofouling is highly complex, involving dust, wax, pollen, microorganisms, and pesticide residues, and a single wettability-based strategy can hardly resist all these foulants simultaneously. In addition, thick or non-breathable coatings may compromise biocompatibility, conformal attachment, and signal transduction, while strong anti-fouling properties often trade off against sensing sensitivity and response time. Therefore, future anti-biofouling designs should focus on multifunctional, mechanically robust, and biocompatible interfaces that maintain stable sensing performance under harsh field conditions [303]. Meanwhile, lightweight agriculture-focused AI edge algorithms can simplify CNN and LSTM network architectures for embedding into low-power terminal chips to realize on-site real-time stress early warning without cloud transmission and reduce terminal power load and server data pressure. Finally, roll-to-roll scalable printing manufacturing should replace expensive vacuum deposition processes to cut production costs and enable dense low-cost layout on scattered smallholder farms and large intensive planting bases to accelerate the industrial popularization of agricultural wearable sensing equipment. Nevertheless, a notable gap remains between laboratory performance and large-scale agricultural deployment. Most reported wearable sensors are still validated under controlled laboratory conditions or small-scale field trials, where temperature, humidity, illumination, and mechanical disturbance can be carefully regulated. In real farm environments, however, analytical performance often degrades owing to environmental fluctuations, long-term biofouling, mechanical fatigue, and batch-to-batch variation in nanomaterial synthesis and device fabrication. More importantly, future translation should not simply pursue ultra-high sensitivity, but should achieve a rational cost-performance balance among detection sensitivity, device lifetime, manufacturing cost, calibration frequency, and maintenance requirements. For practical adoption by farmers and agricultural enterprises, wearable sensors must be affordable, durable, easy to operate, and compatible with existing agricultural management systems. Therefore, future research should strengthen standardized performance evaluation under real field conditions, develop low-cost and reproducible fabrication routes, and establish clear cost-performance benchmarks to accelerate the translation of wearable sensing technologies from laboratory prototypes to large-scale agricultural applications.