Search Results (3,381)

To address the low efficiency and high subjectivity of manual inspection of Astragalus slices, as well as the limited fine-grained detection accuracy caused by the visual similarity between the characteristic radial “chrysanthemum heart” texture and minor defects such as insect damage and mold, this study proposes a lightweight intelligent detection model named Astra-YOLO. A dataset consisting of 622 original Astragalus slice images from four categories was divided into training, validation, and test sets at a ratio of 8:1:1. Data augmentation was applied exclusively to the training set, resulting in a total of 3110 images. Based on YOLOv11n, three targeted improvements were introduced: GhostConv lightweight convolution was employed to reduce model parameters and computational cost; the parameter-free SimAM attention mechanism was integrated to suppress interference from complex textures and enhance defect feature representation; and Wise-IoU v3 was adopted to improve bounding box regression for precise localization of small defects. The experimental results demonstrate that Astra-YOLO achieves superior performance with only 2.53 million parameters and 6.20 GFLOPs. The model attains an mAP@0.5 of 92.7%, an mAP@0.5:0.95 of 73.8%, a precision of 92.4%, and a recall of 92.1%. These results indicate that Astra-YOLO effectively balances lightweight design and detection accuracy, outperforming the baseline model and other improved variants, thereby providing reliable technical support for industrial online inspection and automated quality grading of Astragalus slices. Full article

Under the dwarf-rootstock dense planting method, existing apple detection models for intelligent harvesting suffer from excessive parameter counts that hinder deployment on resource-constrained devices, while lightweight alternatives often sacrifice detection accuracy. To address this dilemma, this paper proposes YOLO-SPM, a lightweight apple detection model based on the YOLOv12n architecture, specifically designed for complex orchard environments. The core innovation lies in a problem-driven, three-stage collaborative optimization strategy: first, PConv is introduced to replace standard convolutions in the A2C2f module, reducing computational redundancy by exploiting channel-wise feature similarity of apple targets; second, the parameter-free SimAM attention mechanism is embedded in the neck network to enhance the model’s focus on occluded fruit features without increasing model size, while MBConv is integrated into the detection head to further reduce computational cost; third, WIoU v3 is adopted as the loss function to compensate for the accuracy loss incurred by lightweight design through its dynamic focusing mechanism on difficult samples. This complementary design ensures that each module addresses a distinct bottleneck of the native YOLOv12n in orchard scenarios, achieving a balance between efficiency and accuracy rather than simple module stacking. Experimental results demonstrate that YOLO-SPM achieves a precision of 92.8% and mAP@0.5 of 93.1%, outperforming the baseline by 4.8 and 5.3 percentage points, respectively, while reducing parameter count, FLOPs, and memory footprint by 40.2%, 35.4%, and 41.8%. This study provides a feasible solution for high-precision apple identification in dwarf-rootstock dense planting orchard environments, with the potential for integration into automated harvesting systems upon future on-device validation. Full article

Open-vocabulary object detection (OVD) fails at attribute-level discrimination: when instances share a class label yet differ in color, material, or texture, category names provide no appearance-specific cues. Prior attempts to enrich text inputs with LLM-generated descriptions are limited by single-model distribution bias, producing coverage gaps and unstable attribute quality. We propose a Concept Expander framework built on cooperative multi-LLM game theory. Three heterogeneous LLMs generate candidate attributes in parallel; a cooperative Nash equilibrium then selects the final subset by maximizing each model’s minimum utility gain, jointly enforcing semantic quality and cross-source diversity without amplifying any single model’s bias. The resulting Concept Repository contains approximately 5000 discriminative visual priors. A lightweight retrieval module injects the top-k matched attributes into region-level visual features via residual fusion, preserving CLIP’s pretrained alignment while enriching instance representations with fine-grained semantic priors. A semantic consistency loss anchors enhanced features to ground-truth class semantics throughout training. On LVIS, rare-category APr rises from 22.2 to 28.5; on RefCOCO, attribute-conditioned localization accuracy reaches 54.8, confirming that structured multi-LLM semantic priors improve discrimination across long-tail and high-confusion benchmarks. Full article

This paper proposes a novel method for automatically designing lightweight Convolutional Neural Network (CNN) architectures. (1) Background: Automated remote sensing for vegetation monitoring faces challenges from structural complexity and cluttered backgrounds. For detecting parasitic Phoradendron velutinum infestations, existing vision frameworks rely on handcrafted, overparameterized CNNs, limiting deployment on localized edge computing platforms. (2) Methods: To address this efficiency-accuracy trade-off, a two-phase hybrid multi-objective Neural Architecture Search (NAS) strategy is implemented. First, the Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) minimizes classification error and the number of trainable parameters. Second, an Iterated Local Search (ILS) metaheuristic refines promising non-dominated solutions. The approach was evaluated using cost-effective aerial RGB imagery, processing a balanced dataset of 5000 patches ($64\times 64$ pixels) under a rigorous three-way data partition to prevent data leakage. (3) Results: The discovered 10-layer CNN topology achieved high feature-extraction efficiency. On the unseen testing set, the model yielded an Accuracy and F1-Score of $0.979$, a Precision of $0.982$, a Recall of $0.976$, and a Jaccard Index of $0.958$, outperforming the compared models. Operating with only 2040 trainable parameters, the optimized architecture establishes a highly viable paradigm for real-time digital image processing on hardware-constrained monitoring devices. Full article

Morel detection in agroforestry cultivation scenes remains challenging because soil-background camouflage, illumination variation, and dense clustered growth can lead to missed small targets and false positives in background regions. This study proposes Morel-YOLO, a lightweight morel detection method based on YOLOv13n for agricultural perception. The model retains the original multi-scale feature-fusion framework and introduces three targeted modifications: a StarNet backbone for reducing redundant computation, a DSC3k2_DWRSeg module in the shallow P${}_3$ branch for strengthening fine-grained texture and small-target representation, and a Detect_MBConv head for reducing prediction-branch overhead while preserving detection accuracy. On the test set, Morel-YOLO achieves 91.9% precision, 86.6% recall, 93.6% mAP${}_{50}$, and 70.8% mAP${}_{50--95}$, improving mAP${}_{50--95}$ by 1.3 percentage points over YOLOv13n. The model contains 1.48 M parameters, has a model size of 3.31 MB, and requires 6.2 GFLOPs. On the Small-hard and Dense-hard subsets, mAP${}_{50--95}$ reaches 69.1% and 66.8%, respectively, corresponding to gains of 1.5 and 1.3 percentage points over the baseline. Under IoU = 0.75, both false positives and false negatives are also reduced on the two hard subsets. These results suggest that Morel-YOLO improves the balance among detection accuracy, robustness, and model compactness on the evaluated dataset; however, its practical deployment on embedded agricultural platforms still requires dedicated on-device validation. Full article

Water level monitoring is closely linked to the safety of production and daily activities along riverbanks, making real-time and high-precision water level measurement an urgent technical demand. The feature extraction backbone of the Unet model is modified, and the lightweight MobileNet V2 network is adopted in this paper. The constructed network achieves significantly higher computational efficiency than standard convolutions, effectively overcoming the limited real-time performance of conventional water level measurement methods. Furthermore, the coordinate attention (CA) mechanism is integrated into the skip connections of Unet to strengthen the network’s capability to extract key features for water level segmentation, thereby further improving the accuracy of water level detection. A novel piecewise linear fitting method for water level line measurement based on monocular vision is proposed, and field-measured water level data are adopted to verify the calculation results. The main achievements of the improved model include the following: (1) Compared with the baseline model, the improved model MCUnet (MobileNet V2 + CA + Unet) achieves a 5.77% increase in accuracy and a 25.71% improvement in inference speed on the experimental water surface recognition dataset. (2) Taking the field-observed water level as the reference, the mean absolute error of the proposed image-based water level monitoring method reaches approximately 1.69 cm. (3) In comparison with DeepLab, U2net and Unet, the MCUnet model gains accuracy improvements of 4.47%, 2.81% and 5.77% respectively, with the detection frame rate increased by 12 FPS, 15 FPS and 11 FPS correspondingly. Through this work, the paper can provide some theoretical support and technical references for overcoming the limitations of conventional water level measuring devices, including strict installation requirements, limited measurement precision, high deployment and maintenance costs, and cumbersome data processing. Full article

Dense underwater aquaculture poses significant challenges for intelligent image processing because asymmetric occlusion, turbidity, aeration-like bubbles, and motion blur frequently degrade fish contours and quasi-periodic scale textures. These disturbances often cause conventional detectors to miss detections, merge bounding boxes, experience feature collapse, and exhibit unstable counting. To address this problem, we propose DenseFish-v13, a symmetry-aware NMS-free YOLOv13-Mamba framework for dense underwater fish detection and bio-kinematic behavior recognition. The framework integrates a Bio-Harmonic Frequency Gate to preserve biological texture patterns while suppressing bubble-like frequency noise, a Bi-directional Multi-scale Wavelet Mamba backbone for global occlusion-aware structure recovery, and an asymmetry-aware density repulsion strategy to separate highly overlapping fish instances during bipartite matching. In addition, a lightweight Bio-Kinematic Behavior Head converts continuous detections into interpretable trajectory descriptors for behavior-state recognition. Experiments on the Dense-Aqua benchmark, constructed from public aquaculture datasets, show that DenseFish-v13 achieves 64.8% mAP@50:95 and a Counting MAE of 3.7 on the overall test set, while reaching 64.2% mAP@50:95 and a Counting MAE of 4.1 on the extreme-density split. Under a strong synthetic bubble perturbation, the model shows only a 1.3 percentage-point drop in mAP and maintains 125 FPS on Jetson Orin NX. These results demonstrate its effectiveness in robust, real-time underwater aquaculture monitoring. Full article

In response to the challenges in UAV-oriented ground small-object localization and detection, including the easy loss of tiny target features, insufficient scale adaptability, severe interference from complex backgrounds, as well as high missed and false detection rates and the inadequate localization accuracy of the conventional YOLOv11n model in such scenarios, this paper takes YOLOv11n as the basic framework and performs systematic optimization from three aspects, network structure, core modules, and feature enhancement, proposing a lightweight small-object-enhanced detection algorithm named YOLOSO for UAV applications. By introducing a P2 high-resolution feature branch with a stride of 4, a four-scale detection structure consisting of P2-P3-P4-P5 is constructed, which reduces the minimum detection stride from 8 to 4 and alleviates the loss of detailed feature information for ultra-tiny targets. A bidirectional “top-down + bottom-up” multi-scale feature fusion strategy is utilized to improve the complementation between deep semantic information and shallow detailed features, while the core modules C3k2SO and C2PSASO are optimized and redesigned, respectively; by adjusting the channel compression ratio (0.25 for shallow modules and 0.75 for deep modules in C3k2SO; 0.25 in C2PSASO), optimizing the convolution kernel configuration (combining 1 × 3 and 3 × 1 convolutions), increasing the number of attention heads (from 4 to 8), and introducing residual connections with a 1 × 1 convolutional branch, the refinement and focusing ability of small-object feature extraction are improved. Additionally, an Enhanced Dual-branch Convolutional Block Attention Module (ED-CBAM) is proposed to further suppress background interference. Experimental results on the VisDrone2019-DET dataset demonstrate that the proposed YOLOSO contains 3.56M parameters and maintains a lightweight structure, attaining P, R, and mAP50 values of 47.2%, 36.8%, and 37.3% in the test set, which are 4.5 percentage points, 4.8 percentage points, and 3.7 percentage points higher than those of the baseline YOLOv11n (42.7%, 32.0% and 33.6%), respectively. Meanwhile, the medium-to-large version YOLOSO-S (14.85M parameters, 45.3% mAP50) reduces the number of parameters by 53.6% compared with the same-scale Rtdetr-L (32.0M) while achieving significantly better performance (37.8% mAP50). Experiments on the DOTAv1 dataset further confirm the generalization of YOLOSO, achieving 62.2% precision and 27.3% mAP50, outperforming all compared YOLO models. Evaluated on the DOTA-v1 dataset, YOLOSO achieves a feasible FPS of 20.53. Although slightly slower than mainstream lightweight YOLO models, the substantial accuracy gains fully offset the minor inference speed loss, and such performance trade-off is acceptable for practical UAV deployment. Ablation experiments verify that structural optimization (2.8 percentage points mAP50 improvement, from 33.6% to 36.4%) and the proposed C2PSASO (0.7 percentage points mAP50 improvement to 34.3%) and C3k2SO (1.4 percentage points mAP50 improvement to 35.0%) modules all contribute positive performance gains with favorable complementarity. While retaining lightweight characteristics, the model effectively enhances the detection accuracy of small objects in unmanned aerial vehicle scenarios and can provide technical references for practical applications such as remote sensing monitoring and security patrolling. Full article

The Internet of Medical Things (IoMT) enables real-time monitoring and decision support systems in healthcare. However, due to their heterogeneous structure, limited resources, and high criticality, IoMT networks are vulnerable to cyberattacks. This situation increases the need for low-latency, high-accuracy, and generalizable attack detection systems. In this experimental study, the Multi-Scale Depthwise Channel Attention Network (MSCA-Net) model is proposed for multi-class attack detection in IoMT environments. The model consists of three core components: multi-scale depthwise separable convolutions to capture traffic patterns across different time scales, a squeeze-and-excitation-based channel attention mechanism that adaptively weights discriminative features, and a lightweight unidirectional LSTM layer that models temporal dependencies. This architecture enables effective representation learning with low parameter costs. The proposed model was evaluated on the WUSTL-EHMS-2020 and CICIoMT2024 datasets. On the CICIoMT2024 dataset, it achieved 99.75% accuracy and a weighted F1 score of 99.77% in a 6-class scenario. It has also demonstrated competitive results in 19-class fine-grained classification. Experimental comparisons show that MSCA-Net offers a better performance-to-cost trade-off compared to nine different baseline models. Furthermore, it demonstrates a speed advantage of up to two times in inference time. The results obtained at the conclusion of the experimental study demonstrate that the proposed approach effectively addresses the challenges of multi-scale feature extraction, class imbalance, and computational efficiency. Furthermore, the model appears to offer a viable solution for real-time attack detection in IoMT environments. Full article

Rice leaf diseases severely threaten global food security, and efficient on-site detection remains challenging for resource-constrained field inspection robots. This work introduces a lightweight visual detection framework designed for the real-time and accurate identification of rice leaf diseases on agricultural edge mobile platforms. A dataset of 4622 annotated images compiled from mobile-device acquisition and publicly available online sources, covering three representative disease categories, together with an independent public benchmark, was used for evaluation. The framework integrates three complementary modules: adaptive multi-scale feature extraction via a dynamic hybrid convolution backbone (C3k2-DICN), cross-scale parameter sharing in the detection head (CSDH) to reduce redundancy, and dual-path downsampling (ADown) to preserve disease-discriminative information during resolution compression. Compared to the YOLO11n baseline, the proposed approach reduced GFLOPs by 36.5% and parameter count by 34.6%, while achieving 88.42% mAP@0.5 and 45.82% mAP@0.5:0.95 on the compiled dataset and 91.71% mAP@0.5 on the public benchmark, indicating accuracy competitive with or superior to all evaluated comparison models. Deployed on an NVIDIA Jetson TX2 with TensorRT FP16 acceleration, the model ran in real time on-device, reaching 32.2 FPS for the TensorRT inference stage and 19.8 FPS for the full end-to-end pipeline including image pre- and post-processing. The framework offers a practical basis for lightweight on-device rice disease detection; closed-loop validation on a moving field robot is left to future work. Full article

The increasing demand for intelligent transportation systems (ITS) and advanced driver assistance system (ADAS) significantly demands a real-time and robust perception to recognize road-side obstacles in varying different weather settings. This paper presents DriveEdgeAI, a lightweight YOLO11 based embedded deep learning framework for efficient road anomaly detection with the emphasis on potholes, speed bumps and relevant traffic sign detection. We have prepared a custom dataset consisting of 17,061 annotated images to train and test the model under different lighting conditions, weather conditions, and roads configurations. The proposed system also managed to demonstrate good convergence and generalization with a precision@50 of 95.8%, recall@50 of 89.7%, mAP@50 of 95.4%, surpassing previous YOLO versions. The stability and robustness of the model at different thresholds were also substantiated by Precision-Recall and F1-Confidence analyses. DriveEdgeAI was also deployed on a number of edge devices, such as Jetson Nano, Raspberry Pi 5, Intel Movidius VPU and Hailo-8L NPU respectively reaching 9.5 FPS/W and 28.5 FPS for the Raspberry Pi 5 + Hailo-8L version. From these results, one can conclude that DriveEdgeAI is an energy-efficient and scalable solution for real-world ADAS applications. Full article

Sleep-related respiratory disturbances are difficult to monitor continuously outside specialized laboratories because conventional polysomnography is resource-intensive and intrusive. This study presents a contactless edge-AI engineering prototype for detecting controlled voluntary respiratory-motion suppression and motion artifacts using a 60 GHz frequency-modulated continuous-wave radar. The system integrates a 60 GHz radar front end, lightweight local preprocessing, an INT8 one-dimensional convolutional neural network deployed on the Analog Devices MAX78000 CNN accelerator (Analog Devices Thailand, Chon Buri, Thailand), and an event-driven Raspberry Pi Zero 2W gateway for alert transmission. Evaluation was performed using a controlled healthy-volunteer dataset consisting of normal breathing, voluntary breath-holding-induced respiratory suppression, and deliberate motion artifact. The final valid test set contained 270 technically valid 30 s windows balanced across the three classes. The INT8 model achieved an overall accuracy of 92.6% (95% confidence interval: 88.8–95.2%), with a macro-averaged precision, recall, and F1-score of 92.6%, 92.6%, and 92.5%, respectively. Active CNN inference on the MAX78000 consumed 0.152 ± 0.011 mJ and was completed in 5.20 ± 0.11 ms, corresponding to approximately 280-fold lower active inference energy than Python 3.14.6/TensorFlow Lite 2.21.0-based execution on the Raspberry Pi Zero 2W. These results demonstrate the feasibility of privacy-aware, low-power respiratory-pattern classification at the edge. However, the study should be interpreted strictly as an engineering proof-of-concept based on controlled voluntary breathing and movement tasks in healthy volunteers. It is not a clinically validated apnea or obstructive sleep apnea detection system and did not include polysomnography, oxygen saturation measurement, airflow sensing, sleep staging, or diagnosed patient cohorts. Full article

Rapid and non-destructive detection of nitrogen content in greenhouse cucumber leaves is essential for precision fertilization, yet traditional chemical methods are destructive and time-consuming, and existing spectral technologies suffer from high cost and poor field adaptability. This study aims to propose a high-precision detection scheme for cucumber leaf nitrogen content based on a lightweight model, suitable for complex scenarios. A total of 698 cucumber leaf images covering three growth stages were collected to build a segmentation dataset. Four categories and eight types of deep learning segmentation models were optimized and compared, and the optimal one was selected to extract leaf regions. Nine color features were extracted and combined with Kjeldahl-measured nitrogen content to construct and optimize three machine learning models, forming a deep learning segmentation–color feature extraction–machine learning prediction process. The results showed that YOLOv11n achieved the best segmentation accuracy, with an IoU of 0.9212 and AP of 0.9998 for high-resolution images. The optimized XGBoost had the highest prediction accuracy, with an MAE of 0.469, MSE of 0.461, and RMSE of 0.679, which are 10.15%, 8.71%, and 4.36% lower than Support Vector Regression with Radial Basis Function kernel (SVR_RBF) respectively, and its predicted nitrogen content aligned well with true values. The proposed scheme integrating YOLOv11n and XGBoost offers a lightweight technical solution for nitrogen nutrition diagnosis and precise fertilization of greenhouse cucumbers. Full article

Real-time fruit maturity detection in unstructured orchards remains challenging because of variable illumination, fruit occlusion, complex backgrounds, and the limited computing capacity of edge devices. To address these challenges, this study proposes WSS-YOLO, a lightweight detection framework based on YOLOv11n for quince maturity detection. The model introduces WaveletPool to reduce texture loss during downsampling, adopts a GSConv-based Slim-neck to improve feature fusion with lower computational cost, and integrates SimAM to enhance discriminative fruit-region responses without adding trainable parameters. Experiments on a multi-scenario quince maturity dataset show that WSS-YOLO achieves 86.4% precision, 87.5% recall, and 93.4% mAP@0.5, improving the YOLOv11n baseline by 2.3, 1.7, and 2.5 percentage points, respectively. The model contains only 2.23 M parameters and requires 4.1 G FLOPs. Deployment on the NVIDIA Jetson Orin Nano achieved a real-time speed of 23.0 FPS, suggesting a favorable trade-off between detection accuracy and computational efficiency under the tested conditions. Full article

Background: Accurate assessment of neonatal sleep is critical for monitoring brain development and identifying potential neurological disorders, yet manual scoring of multi-channel EEG recordings is labor-intensive and prone to variability. Methods: To address this, we propose a lightweight temporal–spatial feature fusion network for automatic neonatal sleep staging. The model employs a dual-branch architecture to separately capture temporal dependencies and spatial correlations in EEG signals, which are then integrated through feature concatenation and a compact classifier to obtain comprehensive feature representations while maintaining low computational complexity. Results: The framework was evaluated on a clinical neonatal dataset (CHFD) for tasks including sleep–wake classification, quiet sleep detection, and three-stage sleep staging, achieving superior performance compared with several state-of-the-art methods. Additional evaluation on the MASS-S3 adult dataset demonstrate that the model retains competitive accuracy and F1-score, indicating strong generalization across populations. Conclusions: These results suggest that jointly modeling temporal and spatial features enables robust and efficient automatic sleep staging. The proposed approach offers a practical solution for clinical applications and edge deployment, providing reliable, multi-dimensional assessment of neonatal brain activity and laying the groundwork for future studies integrating larger datasets or multimodal physiological signals. Full article