Identification of Cotton Defoliation Sensitive Materials Based on UAV Multispectral Imaging

Abstract

(1) Background: This study aims to analyze the defoliation and boll opening performance of 123 upland cotton germplasm resources after spraying defoliant, using multispectral data to select relevant vegetation indices and identify germplasm resources sensitive to defoliants, providing methods for cotton variety improvement and high-quality parental resources. (2) Methods: 123 historical upland cotton germplasm resources from Xinjiang were selected, and the defoliation and boll opening of cotton leaves were investigated at 0, 4, 8, 12, 16, and 20 days after defoliant application. Simultaneously, multispectral digital images were collected using drones to obtain 12 vegetation indices. Based on defoliation rate, the optimal vegetation index was selected, and defoliant-sensitive germplasm resources were identified. (3) Results: The most significant difference in defoliation rate of cotton germplasm resources occurred 16 days after application. Cluster analysis grouped the 123 breeding materials into three categories, with Class I showing the best defoliation effect. Among the 12 vegetation indices, the Plant Senescence Reflectance Index (PSRI) has the highest correlation coefficient with the defoliation rate; and when the PSRI value is higher, the defoliation effect of the material is better. By comparing the traditional investigation method with the unmanned aerial vehicle multispectral technology, 15 cotton materials sensitive to defoliants were determined, with a defoliation rate of over 85%, a lint percentage ranging from 76.67% to 98.04%, and a PSRI value ranging from 0.1607 to 0.1984. (4) Conclusions: The study found that the vegetation index with sensitive response can be used as an effective indicator to evaluate the sensitivity of cotton breeding materials to defoliants. Using an unmanned aerial vehicle (UAV) equipped with vegetation indices for screening shows a high consistency with the manual investigation and screening method in screening excellent defoliation materials; it proves that it is feasible to screen cotton breeding materials with excellent defoliation effects using UAV multispectral technology.

1. Introduction

Cotton is an important global economic crop. Xinjiang is the largest cotton-producing area in China. In 2024, the national cotton output was 6.164 million tons. Of this total, the cotton output in Xinjiang was 5.686 million tons, accounting for 92.2% of the national total cotton output [1]. With rising costs and the development of agricultural mechanization, the area planted with machine-harvested cotton in Xinjiang has expanded, forming a new planting model that integrates equipment and technology. The mechanization rate of plowing, sowing, and harvesting has reached 97%, effectively ensuring a high cotton yield [2]. During the cotton harvesting stage, there are issues such as increased labor costs, low harvesting efficiency, and extended harvesting time due to labor shortages [3]. With the development of mechanized cotton-picking technology, machine-picked cotton has greatly alleviated the labor pressure.The mature mechanized cotton-picking technology has promoted the rapid development of the whole-process mechanization and precision agriculture of cotton production in China. Popularizing machine-picked cotton has become the trend of the future development of the cotton industry, and chemical defoliation and ripening is one of the key technologies to ensure the smooth completion of machine-picked cotton [4].

Currently, agricultural monitoring is gradually achieving an integrated approach that combines space, air, and ground, focusing on comprehensive monitoring of resources, environment, and crop growth status, aimed at providing strong support for the smart development of agriculture. Unmanned aerial vehicle (UAV) remote sensing, with its advantages of short cycle, simple operation, dynamic data acquisition and low cost [5], has become a new trend and research hotspot in agricultural monitoring. In recent years, remote sensing technology has been applied in the inversion research of physiological and biochemical parameters such as crop biomass [6], chlorophyll content [7], and water and nitrogen content [8]. In terms of characterizing leaf changes, spectral reflectance and vegetation indices have their own characteristics. The spectral reflectance index based on the green and red wavelength regions can more precisely reveal the continuous changes in pigment composition and its quantity during the leaf senescence process [9]. Zhang [10] used the spectral indices and texture features extracted from the UAV remote sensing image and cotton FPAR measurement data to construct a machine learning model, and evaluated the performance of the model.

In the early stages of cotton harvesting, defoliants are used to promote ripening and address the problem of late-maturing cotton with excessive green growth. Scientific and reasonable field management measures, along with the correct use of defoliants, are fundamental to achieving high yield and quality cotton [11]. During the mechanized cotton harvesting process, the application of defoliants for defoliation and ripening is an indispensable key step. The innovation of machine-harvested cotton germplasm resources and the breeding of new varieties are critical for the development of the industry. Defoliant sensitivity, as an important breeding index, has placed higher demands on breeding work. In the study of cotton defoliation and ripening, the evaluation of defoliation effect mainly relies on the traditional manual fixed-point investigation [12], which is time-consuming, inefficient and the results are accidental.In contrast, unmanned aerial vehicle remote sensing has the advantages of high resolution and real-time monitoring, and its application in agricultural monitoring is increasingly widespread [13], which provides new thinking and research directions for the monitoring of cotton defoliation rate and material screening. Yan et al. [14] used a quadcopter UAV to collect multispectral images of cotton before and after spraying defoliants, calculated vegetation indices, and extracted cotton leaf coverage information. Ma et al. [15] used UAV RGB images as the data source to extract 14 different visible light vegetation indices, then constructed a fast and accurate cotton defoliation rate monitoring model. Ma et al. [16] found that RF is more effective in screening features to monitor defoliation and boll opening rates. Using the model results, a comprehensive evaluation index of defoliation effect can be constructed, and the defoliation effect index can quickly determine the harvest time of cotton. Based on the previous research results, due to the characteristics of UAV images being easy to obtain, low cost, and relatively simple processing process, coupled with the high applicability of related vegetation indices, it provides a new idea and direction for using UAV multispectral technology to monitor and screen excellent breeding materials for cotton defoliation effect.

At present, there are relatively few studies on using unmanned aerial vehicle (UAV) remote sensing technology to monitor the defoliation effect of cotton and screen materials. In this experiment, 123 upland cotton germplasm resources after spraying defoliant were screened for defoliation sensitivity, and combined with UAV multispectral data, the feasibility of UAV multispectral technology in monitoring, classifying and screening germplasm resources was evaluated. Moreover, the traditional investigation method and the UAV multispectral technology method were compared to screen out defoliation-sensitive cotton germplasm resources, further improving the application of UAV multispectral technology and vegetation index in the evaluation of cotton defoliation sensitivity, providing a theoretical basis and technical reference for screening cotton germplasm resources with suitable defoliant sensitivity for mechanized harvesting.The main contributions of this study are as follows:

(1)

The vegetation indices selected, which are related to cotton defoliation, can serve as evaluation indicators for assessing the effectiveness of cotton defoliation, providing a reference for subsequent research.

(2)

A framework for evaluating materials using drone-based multispectral technology has been established, offering theoretical foundations and technical references for selecting cotton germplasm resources with superior defoliation effects suitable for mechanized harvesting.

(3)

The effectiveness of cotton defoliation was assessed and selected using drone-based multispectral technology, and a comparison and analysis was conducted with traditional manual screening of cotton materials sensitive to defoliation. The results indicate that the use of drone-mounted vegetation indices for screening is highly consistent with the manual survey and selection methods in identifying defoliation-effective materials. This further enhances the application of drone multispectral technology and vegetation indices in evaluating cotton defoliation sensitivity.

2. Materials and Methods

2.1. Materials and Geographical Location

In this experiment, 123 cotton germplasm resources were selected, all of which were collected and provided by the Key Laboratory of Crop Breeding and Biotechnology of Xinjiang (

Table 1. Names of 123 cotton breeding materials used for testing.

No.	Name	Source	No.	Name	Source	No.	Name	Source	No.	Name	Source	No.	Name	Source
YT001	Liaomian 25	The Yellow River Basin	YT028	Xinluzao 74	Northwest Inland	YT053	Ari971	Abroad	YT078	Simian 2	The Yangtze River Basin	YT103	Source Cotton 8	Northwest Inland
YT002	Liaomian 35	The Yellow River Basin	YT029	Xinluzao 75	Northwest Inland	YT054	BP52	Abroad	YT079	Simian 3	The Yangtze River Basin	YT104	J206-5	Northwest Inland
YT003	Shengmian 6	Other Sources	YT030	Xinluzao 76	Northwest Inland	YT055	Si-1470	Abroad	YT080	Xinluzao 42	Northwest Inland	YT105	Guanmian 678	Self-bred Promotion
YT004	Xinmian 3	Northwest Inland	YT031	Xinluzao 78	Northwest Inland	YT056	J02-247	The Yangtze River Basin	YT081	Xinluzao 33	Northwest Inland	YT106	Baijin 3045	Self-bred Promotion
YT005	Xinshi K18	Northwest Inland	YT032	Xinluzao 79	Northwest Inland	YT057	Z37less	Abroad	YT082	Xinluzao 23	Northwest Inland	YT107	Guanmian 614	Self-bred Promotion
YT006	Xinshi K24	Northwest Inland	YT033	Xinluzao 84	Northwest Inland	YT058	Bamian 1	The Yangtze River Basin	YT083	Xinluzao 10	Northwest Inland	YT108	Kang 41	Self-bred Promotion
YT007	Chuangmian 512	The Yellow River Basin	YT034	Xinluzhong 38	Northwest Inland	YT059	Changkangmian	The Yangtze River Basin	YT084	Xinluzao 8	Northwest Inland	YT109	Feng Haimian	Self-bred Promotion
YT008	Longmian 10	Northwest Inland	YT035	Xinluzhong 50	Northwest Inland	YT060	Chuan 169-6	The Yangtze River Basin	YT085	Tu 83-161	Northwest Inland	YT110	Fengze 7	Self-bred Promotion
YT009	Jinken 1441	Northwest Inland	YT036	Mutant1	Self-bred Materials	YT061	Jingzhou Degenerated Cotton	The Yangtze River Basin	YT086	Xinluzao 47	Northwest Inland	YT111	Huimin 52	Self-bred Promotion
YT010	Jinken 1565	Northwest Inland	YT037	Mutant2	Self-bred Materials	YT062	Jing 55173	The Yellow River Basin	YT087	Xinluzao 48	Northwest Inland	YT112	Huimin 4	Self-bred Promotion
YT011	Jinken 1643	Northwest Inland	YT038	Mutant3	Self-bred Materials	YT063	Jinmian 36	The Yellow River Basin	YT088	Xinluzao 49	Northwest Inland	YT113	Guanmian V5	Self-bred Promotion
YT012	Jiumian NE01	Other Sources	YT039	Mutant4	Self-bred Materials	YT064	Jinzimian King	Abroad	YT089	Xinluzao 52	Northwest Inland	YT114	Genesis 8	Self-bred Promotion
YT013	W8225	The Yellow River Basin	YT040	Mutant5	Self-bred Materials	YT065	Jiangsu Cotton 1	The Yangtze River Basin	YT090	Xinluzao 61	Northwest Inland	YT115	Hexin Seed Industry 14	Self-bred Promotion
YT014	Xinniumian 206	Other Sources	YT041	Mutant6	Self-bred Materials	YT066	Jimian 8	The Yellow River Basin	YT091	Xinluzhong 6	Northwest Inland	YT116	Guanmian 648	Self-bred Promotion
YT015	Zhongmiansuo 115	The Yellow River Basin	YT042	Mutant7	Self-bred Materials	YT067	Jijiaohongye Mian	Abroad	YT092	Xinluzhong 14	Northwest Inland	YT117	Genesis 5	Self-bred Promotion
YT016	Xinluzao 27	Northwest Inland	YT043	Mutant8	Self-bred Materials	YT068	Han 241	The Yellow River Basin	YT093	Xinluzhong 36	Northwest Inland	YT118	Zhongya Huijin 6	Self-bred Promotion
YT017	Xinluzao 50	Northwest Inland	YT044	Mutant9	Self-bred Materials	YT069	Ganmian 12	The Yangtze River Basin	YT094	Xinluzhong 41	Northwest Inland	YT119	Fengdekang 4	Self-bred Promotion
YT018	Xinluzao 51	Northwest Inland	YT045	Mutant10	Self-bred Materials	YT070	Ferganskaya 175	Abroad	YT095	Xinluzhong 54	Northwest Inland	YT120	Genesis 7	Self-bred Promotion
YT019	Xinluzao 54	Northwest Inland	YT046	R22-46	Self-bred Materials	YT071	Miaohua in Judian Township, Lijiang County, Yunnan	The Yangtze River Basin	YT096	Zhongmiansuo 17	The Yellow River Basin	YT121	Genesis 8	Self-bred Promotion
YT020	Xinluzao 55	Northwest Inland	YT047	Xinluzao 11	Northwest Inland	YT072	Daihongdai	The Yangtze River Basin	YT097	Zhongmiansuo 12	The Yellow River Basin	YT122	Genesis 3	Self-bred Promotion
YT021	Xinluzao 57	Northwest Inland	YT048	Zhongmian Institute 43	The Yellow River Basin	YT073	Kuche 96515	Northwest Inland	YT098	Zhong 203016	The Yellow River Basin	YT123	Xiangsui Seed Industry 2	Self-bred Promotion
YT022	Xinluzao 60	Northwest Inland	YT049	70-1437	The Yangtze River Basin	YT074	Liaomian 9	The Yellow River Basin	YT099	Yuan 247-31	The Yellow River Basin	YT124	Jike Huayu 1	Self-bred Promotion
YT024	Xinluzao 64	Northwest Inland	YT050	73-184	The Yellow River Basin	YT075	Zhongmiansuo 23	The Yellow River Basin	YT100	Yumian 1	The Yellow River Basin	YT125	Xinluzao 73	Northwest Inland
YT025	Xinluzao 68	Northwest Inland	YT051	AC321	Abroad	YT076	Shaan 416	The Yellow River Basin	YT101	Xinluzhong 68	Northwest Inland
YT026	Xinluzao 69	Northwest Inland	YT052	Ari3697	The Yellow River Basin	YT077	Shen 547	The Yangtze River Basin	YT102	Xinluzhong 75	Northwest Inland

). These materials include 48 from the Northwest Inland Cotton Area, 19 from the Yellow River Basin Cotton Area, 14 from the Yangtze River Basin Cotton Area, 8 from overseas, 20 from self-bred materials by Xinjiang regional seed companies, 11 from self-bred materials by the research team, and 3 other materials. The experiment was conducted in 2024 at the Huyanghe City Experimental Base of the Agricultural Science Research Institute of the Seventh Division of the Xinjiang Production and Construction Corps (Latitude 44°20′–47°04′, Longitude 83°51′–85°51′). The base has a continental arid climate with long sunshine hours, scarce precipitation, high evaporation, a frost-free period of 170 days, and an average annual temperature of 11 °C. The temperature change in September is shown in Figure 1. The average temperature is approximately 20 °C, the minimum temperature is 9 °C, and the maximum temperature is 31 °C.

2.2. Experimental Design and Treatments

The experiment was designed with two treatments: one with defoliant spraying (DT) and one without defoliant spraying (CK), each with two replicates. Protective rows were also set up. The plot length was 2 m, with 30 membranes per zone, and each membrane covering three rows. Mechanical mulching with hole punching and manual seeding were applied. One membrane was planted with one cotton material. Drip irrigation was applied beneath the film, and other field management was the same as in the field. The defoliant was a 540 g/L thiabendazole diuron suspension (containing 180 g/L diuron and 360 g/L thiabendazole) and a special additive (This defoliant is produced and provided by Jiangsu Ruibang Agrochemical Co., Ltd.). On 5 September, the defoliant treatment was carried out. When the treatment group (DT) unmanned aerial vehicle sprayed the defoliant for the first time, 13–15 mL of 540 g/L thidiazuron + diuron suspension concentrate was used per mu, with an additive added at a ratio of 1:4. The dosage of ethephon was 70–100 mL per mu, and it was sprayed with 30–40 L of water. On 13 September, the defoliant was sprayed for the second time, and 10–12 mL of 540 g/L thidiazuron + diuron suspension concentrate was used per mu, with an additive added at a ratio of 1:4. The control group (CK) did not spray the defoliant but sprayed the same amount of clear water.

2.3. Investigation Contents and Measurement Methods

2.3.1. Field Data Acquisition in Datian

Five consecutive and uniformly growing cotton plants were selected in each plot for tagging. According to “Descriptors and Data Standard for Cotton Germplasm Resources” edited by Du Xiongming et al. [17], the agronomic traits were investigated. Before and after the spraying of defoliant, the number of cotton plant leaves and boll opening numbers of the CK and DT groups at the fixed points and plants were investigated on the 0th day of spraying defoliant and 4 days, 8 days, 12 days, 16 days, and 20 days after spraying defoliant. When conducting the investigation of the ground defoliation rate, cotton leaves smaller than 2 cm² were not calculated. When conducting the last field data investigation, the number of effective bolls were investigated.

2.3.2. UAV Multispectral Data Acquisition

The DJI Phantom 4 quadcopter multispectral unmanned aerial vehicle was used for the acquisition of digital images. The DJI Phantom 4 Multispectral Edition has 6 1/2.9-inch CMOS sensors, including 1 color sensor for visible light imaging and 5 monochrome sensors for multispectral imaging. The effective pixels of a single sensor are 2.08 million (total pixels are 2.12 million). Before and after spraying the defoliant, a manual investigation was conducted, and simultaneously image acquisition work was carried out. The shooting time period was between 12:30 and 14:00. The flight altitude of the unmanned aerial vehicle was set at 20 m. The DJI GS Pro Ground Station Professional Edition was used to plan the fixed-point flight route for the target cotton field, and the flight speed was set at 1 m/s, with an image overlap rate of 80%. Before the flight, a multispectral sensor was used to shoot the radiation calibration plate to facilitate the subsequent radiation calibration of the images. Digital image data were collected at the same intervals as the manual investigation, with UAV data acquisition performed at 0 d, 4 d, 8 d, 12 d, 16 d, and 20 d after the first spraying of defoliant. The multispectral camera bands and wavelength ranges of the DJI Phantom 4 Multispectral Edition UAV are shown in

Table 2. Multispectral Camera Band Parameters.

Band	Central Wavelength (nm)	Bandwidth (nm)
blue	450	16
green	560	16
red	650	16
red_edge	730	16
nir	840	26

.

2.4. Data Processing

The data were statistically analyzed using EXCEL 2010, IBM SPSS Statistics 20 software, GraphPsd Prism 8.0.1 software, and R 4.3.1 software; Pix4Dmapper 4.4.12 software was used for UAV digital image stitching and preprocessing, and the ArcGIS 10.2 software platform was used for extracting multispectral image data.

2.4.1. Calculation of Defoliation Rate and Boll Opening Rate

The defoliation rate (RD) was calculated using the following formula:

RD(%) = (N₀ − N₁)/N₀ × 100%,

(1)

In the equation, N₀ is the number of leaves before the defoliant treatment and N₁ is the number of leaves after the defoliant treatment.

The boll shedding rate (RB) was calculated using the formula:

RB(%) = B₁/B₀ × 100%,

(2)

In the equation, B₁ is the number of bolls that are opening and B₀ is the number of effective bolls.

2.4.2. Unmanned Aerial Vehicle Data Processing

The processing of UAV high-definition digital images was carried out using Pix4D software. The UAV digital images containing location information (including latitude and longitude coordinates and altitude information) were imported into Pix4D to complete image stitching. The output result generate a multispectral image. The system automatically performed image stitching according to the settings. Once the stitching was successful, an image quality report was generated, and the image quality was evaluated through the empty three-ray point cloud. The index calculator module was used to generate the reflection map of the test site, and the obtained digital orthophoto map and 5 index map files in TIFF format of different bands were exported and stored (Figure 2).

The ArcGIS software platform was used for analysis. First, a new folder was created for the digital orthophoto map stitched by Pix4D software and 5 index maps in TIFF format of different bands; the digital orthophoto map was imported into ArcMAP 10.2 software to create a pyramid, and the pyramid resampling technique was selected to be the nearest neighbor method; a new Shapefile feature file was created in the working directory, the rectangular surface was used to divide the sampling points in the community, and finally a Shapefile feature file was obtained containing multiple-point rectangular surfaces of the sampling community; the average reflectance of the research area of each cotton material as the reflectance value of the sample were extracted and exported. Finally, the vegetation index was calculated (Figure 3). (Without further dividing the pixel semantics, regardless of the pixel semantics (pixels representing leaf area, cotton bolls, or other things), it was averaged for all pixels within the red box.

2.4.3. Vegetation Index

The vegetation index is a key parameter for evaluating the growth status of crops. Through in-depth discussion and analysis of the existing literature, this study screened out 12 vegetation indices closely related to the leaf abscission characteristics (

Table 3. Summary of vegetation indices.

No.	Vegetation Index	Abbreviation	Formula	Source
1	Normalized Difference Vegetation Index	NDVI	$\left(\mathrm{n}\mathrm{i}\mathrm{r}-\mathrm{r}\mathrm{e}\mathrm{d}\right)/\left(\mathrm{n}\mathrm{i}\mathrm{r}-\mathrm{r}\mathrm{e}\mathrm{d}\right)$	[18]
2	Normalized Green Difference Vegetation Index	GNDVI	$\left(\mathrm{n}\mathrm{i}\mathrm{r}-\mathrm{g}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}\right)/\mathrm{n}\mathrm{i}\mathrm{r}+\mathrm{g}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}$	[19]
3	Transformed Vegetation Index	TVI	$\sqrt{\left(\mathrm{N}\mathrm{D}\mathrm{V}\mathrm{I}+0.5\right)}$	[18]
4	Ratio Vegetation Index	RVI	$\mathrm{n}\mathrm{i}\mathrm{r}/\mathrm{r}\mathrm{e}\mathrm{d}$	[20]
5	Soil-Adjusted Vegetation Index	SAVI	$\left(\mathrm{n}\mathrm{i}\mathrm{r}-\mathrm{r}\mathrm{e}\mathrm{d}\right)\times \left(1+\mathrm{L}\right)/\left(\mathrm{n}\mathrm{i}\mathrm{r}+\mathrm{r}\mathrm{e}\mathrm{d}+\mathrm{L}\right)$	[21]
6	Enhanced Vegetation Index	EVI	$2.5\times \left(\mathrm{n}\mathrm{i}\mathrm{r}-\mathrm{r}\mathrm{e}\mathrm{d}\right)$	[22]
7	Excess Green Minus Red	EXGR	$2\times \mathrm{g}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}-2.4\times \mathrm{r}\mathrm{e}\mathrm{d}$	[23]
8	Modified Chlorophyll Absorption Reflectance Index	MCARI	$\left[\left(\mathrm{red\_edge}-\mathrm{r}\mathrm{e}\mathrm{d}\right)-0.2\times \left(\mathrm{red\_edge}-\mathrm{g}\mathrm{r}\mathrm{e}\mathrm{e}\mathrm{n}\right)\right]\times \left(\mathrm{red\_edge}/\mathrm{r}\mathrm{e}\mathrm{d}\right)$	[24]
9	Modified second ratio index	MSRI	$\left(\mathrm{n}\mathrm{i}\mathrm{r}/\mathrm{r}\mathrm{e}\mathrm{d}-1\right)/\sqrt{\mathrm{n}\mathrm{i}\mathrm{r}/\mathrm{r}\mathrm{e}\mathrm{d}+1}$	[25]
10	Moisture Vegetation Index	MVI	$\sqrt{\left(\mathrm{n}\mathrm{i}\mathrm{r}-\mathrm{r}\mathrm{e}\mathrm{d}\right)/\left(\mathrm{n}\mathrm{i}\mathrm{r}+\mathrm{r}\mathrm{e}\mathrm{d}\right)+0.5}$	[26]
11	Structure Independent Pigment Index	SIPI	$\left(nir-blue\right)/\left(nir-red\right)$	[27]
12	Plant Senescence Reflectance Index	PSRI	$\left(red-blue\right)/nir$	[28]

). These vegetation indices cover multiple aspects from chlorophyll content to vegetation water status, and can reflect the growth status of cotton and its leaf abscission characteristics, thereby providing a scientific basis for the subsequent material screening and variety improvement.

3. Results and Analysis

3.1. Descriptive Statistical Analysis of Phenotypic Traits of 123 Upland Cotton Germplasm Resources

The agronomic traits of 123 upland cotton germplasm resources were statistically analyzed. The results showed that the plant height of the germplasm materials was 59.20–120.80 cm, the height of the first fruiting branch was 5.80–38.00 cm, the number of fruiting branches was 5.80–13.75, the number of effective fruiting branches was 3.60–12.00, and the number of bolls per plant was 4.80–14.20. The coefficient of variation of each trait exceeded 10% (

Table 4. Descriptive statistics of agronomic traits of germplasm resources.

Traits	Average	Standard Deviation	Min	Max	Coefficient of Variation (%)
Plant height (cm)	87.30	9.64	59.20	120.80	11.04
Height of the first fruiting branch (cm)	20.41	4.70	5.80	38.00	23.00
Number of fruiting branches	10.17	1.19	5.80	13.75	11.66
Number of effective fruiting branches	6.58	1.15	3.60	12.00	17.47
Number of bolls per plant	8.19	1.65	4.80	14.20	20.16

). The coefficient of variation and extreme values of the agronomic traits of cotton germplasm materials indicated that there were significant differences in each trait among different materials, revealing the rich genetic variation in the 123 upland cotton germplasm resources, which provided a basis for the comparison and screening of planting materials.

3.2. The Effect of Defoliants on the Defoliation Rate and Boll Opening Rate of Cotton Germplasm Resources

To study the defoliation effect of cotton after the application of defoliants, the defoliation rate after the application of defoliants was statistically analyzed (Figure 4a), and a significant t-test was conducted on the defoliation rate at the corresponding time points (Figure 4b). The results show that the defoliation rate continues to rise with the increase of days after application, indicating that the defoliant effectively promotes the defoliation process of cotton. Extreme value analysis shows that the defoliation effects of 123 germplasm resources are significantly different, and the defoliation rate change in the treatment group at the T₄ period (16 d) is the most significant, providing a basis for the in-depth analysis of subsequent defoliation-sensitive materials.

In addition, the analysis of the change in the boll opening rate of cotton after spraying defoliants (Figure 4c) and the significant t-test of the boll opening rate at different investigation times (Figure 4d) found that the boll opening rate shows an increasing trend over time, and the boll opening rate in the experimental area where defoliants are sprayed increases more rapidly. Increasing the proportion of boll opening is beneficial to the improvement of fiber quality and the increase of yield. Compared with the control CK group, the boll opening rate in the experimental area treated with defoliants increased significantly, indicating that spraying defoliants has a significant promoting effect on the boll opening of upland cotton germplasm resources.

3.3. Screening of Defoliation-Sensitive Varieties Based on Defoliation Rate

The change trend of the defoliation rate of cotton materials after spraying defoliant shows (Figure 4) that a significant difference in the defoliation rate occurs in the T₄ period, which is convenient for screening materials with good defoliation effects. Therefore, a cluster analysis of the defoliation rate in the T4 period (Figure 5) is conducted, and 123 germplasm resources are divided into three categories. Among them, category I belongs to the defoliation-sensitive germplasm resources, with 37 copies (

Table 5. Defoliant-sensitive germplasm resources.

No.	Material Name	Defoliation Rate (%)	Lint Percentage (%)	No.	Material Name	Defoliation Rate (%)	Lint Percentage (%)
YT015	Zhongmiansuo 115	84.76	89.29	YT092	Xinluzhong 14	89.36	61.84
YT031	Xinluzao 78	90.59	86.60	YT093	Xinluzhong 36	88.32	88.31
YT033	Xinluzao 84	87.35	88.24	YT099	Yuan 247-31	92.57	94.20
YT039	Mutant 4	88.03	78.48	YT100	Yumian 1	90.48	84.88
YT047	Xinluzao 11	87.41	91.43	YT101	Xinluzhong 68	92.15	87.04
YT061	Jingzhou Degenerated Cotton	84.80	90.00	YT102	Xinluzhong 75	86.52	88.00
YT065	Jiangsu Cotton 1	88.10	75.29	YT104	J206-5	88.56	91.86
YT066	Jimian 8	84.82	95.35	YT107	Guomian 614	88.31	94.20
YT068	Han 241	85.81	86.32	YT109	Fenghaimian	87.91	86.25
YT069	Ganmian 12	84.92	81.36	YT112	Huimin 4	92.82	96.59
YT072	Daihongdai	89.17	77.17	YT113	Guamian V5	90.97	88.06
YT074	Liaomian 9	85.95	97.08	YT114	Genesis 8	89.47	96.15
YT075	Zhongmiansuo 23	85.16	80.85	YT115	Hexin Seed Industry 14	93.08	93.42
YT076	Shan 416	87.43	76.67	YT116	Guanmian 648	84.95	95.51
YT078	Simian 2	95.12	82.72	YT118	Zhongya Huijin 6	93.05	97.89
YT082	Xinluzao 23	85.28	91.57	YT119	Fengdekang 4	88.11	100.00
YT083	Xinluzao 10	93.43	93.24	YT122	Genesis 3	89.73	90.65
YT087	Xinluzao 48	92.67	81.91	YT125	Xinluzao 73	94.23	98.04
YT091	Xinluzhong 6	87.57	89.81

), accounting for approximately 30%, and the defoliation rate is between 84.76% and 95.12%; category II belongs to the moderately defoliation-sensitive intermediate materials, with 53 copies, accounting for approximately 48%, and the defoliation rate is between 72.22% and 84.10%; category III belongs to the defoliation-insensitive germplasm resources, with 27 copies, accounting for approximately 22%, and the defoliation rate is between 53.50% and 71.34%.

3.4. Screening of Defoliation-Sensitive Materials of Cotton Based on Multispectral

3.4.1. Changes in Multispectral Reflectance Values

Through the stitching and processing of unmanned aerial vehicle (UAV) multispectral images, the values of five bands were extracted, and their variation is shown in Figure 6. It is found that the values of the red, green, and blue bands show an upward trend with the shedding of cotton leaves, while the values of the near-infrared (nir) and red-edge (red_edge) bands show a downward trend with the shedding of cotton leaves. Through the investigation of field experiment data, it is found that there is a significant difference in the defoliation situation at 16 days (T₄) after the application of the defoliant. Therefore, the multispectral values at 16 days (T₄) after the application of the pesticide are selected for the subsequent study.

3.4.2. Analysis of the Correlation Between Multi-Spectral Bands and Vegetation Indexes and Defoliation Rate

A correlation analysis was conducted on the defoliation rate, multispectral bands, and vegetation indices on the 16th day after pesticide application (Figure 7). PSRI shows a significant positive correlation with the defoliation rate, and the highest correlation coefficient reaches 0.52; an increase in the PSRI value indicates an increase in canopy pressure, the beginning of vegetation senescence, and the maturity of plant fruits. This index is commonly used to monitor vegetation health, vegetation physiological stress, and crop yield analysis. The SIPI vegetation index is often used to monitor plant health in areas with highly variable canopy structure or LAI, mainly to identify the early signs of crop diseases or other stress causes. When screening vegetation indices, the correlation between PSRI and defoliation rate is higher than that of SIPI. Therefore, the PSRI vegetation index is finally selected for the next step of analysis. Among them, MSRI, RVI, NDVI, SAVI, TVI, MVI, MCARI, EVI, and GNDVI show a significant negative correlation with the defoliation rate, and the correlation coefficient reaches more than −0.40. These several vegetation indices have a significant correlation with the defoliation rate and can be used as relevant indicators to evaluate the defoliation of cotton.

3.4.3. PSRI Clustering Screening for Defoliation-Sensitive Upland Cotton Germplasm Resources

Relevance analysis found that the correlation between PSRI and defoliation rate is the highest. Therefore, 123 germplasm resources are divided into three categories using the PSRI value (Figure 8). Among them, 24 germplasm resources with excellent defoliation effect are classified as Type I (

Table 6. PSRI screening of germplasm resources sensitive to defoliants.

No.	Material Name	Defoliation Rate (%)	Lint Percentage (%)	PSRI	No.	Material Name	Defoliation Rate (%)	Lint Percentage (%)	PSRI
YT006	New Stone K24	75.34	84.54	0.1696	YT085	Tu 83-161	72.49	92.31	0.1640
YT039	Mutant4	88.03	78.48	0.1795	YT099	Yuan 247-31	92.57	94.20	0.1819
YT040	Mutant5	83.23	73.42	0.1628	YT100	Yumian 1	90.48	84.88	0.1852
YT044	Mutant9	78.34	70.83	0.1662	YT101	Xinluzhong 68	92.15	87.04	0.1723
YT045	Mutant10	79.51	58.46	0.1738	YT102	Xinluzhong 75	86.52	88.00	0.1733
YT047	Xinluzao 11	87.41	91.43	0.1720	YT104	J206-5	88.56	91.86	0.1756
YT058	Bamian 1	75.76	95.10	0.1632	YT107	Guanmian 614	88.31	94.20	0.1607
YT068	Han 241	85.81	86.32	0.1690	YT110	Fengze 7	82.46	96.15	0.1650
YT074	Liaomian 9	85.95	97.08	0.1984	YT111	Huimin 52	79.37	98.57	0.1700
YT076	Shan 416	87.43	76.67	0.1859	YT113	Guanmian V5	90.97	88.06	0.1694
YT078	Simian 2	95.12	82.72	0.1649	YT123	Xiangsui Seed Industry 2	83.62	95.89	0.1697
YT083	Xinluzao 10	93.43	93.24	0.1852	YT125	Xinluzao 73	94.23	98.04	0.1609

), accounting for approximately 19.5%, with PSRI values ranging from 0.1607 to 0.1984; 43 germplasm resources with a general defoliation effect are classified as Type II, accounting for approximately 35.0%, with PSRI values ranging from 0.1358 to 0.1588; and 56 germplasm resources with a poor defoliation effect are classified as Type III, accounting for approximately 45.5%, with PSRI values ranging from 0.0763 to 0.1350.

3.5. Defoliation Rate Classification and PSRI Classification Screening Materials Consistency Evaluation

During the field investigation of the cotton defoliation rate, due to human factors such as the movement and touching of the investigators, the leaves often fall unnaturally, thereby affecting the accuracy of the investigation data. This study employed the traditional manual investigation method of defoliation rate and the classification and screening method of vegetation index PSRI to select cotton materials that are sensitive to defoliants. During the experiment, cluster analysis was conducted through the vegetation index PSRI, and 24 materials with significant defoliation effects were successfully screened out. These were compared with the 37 defoliation-sensitive materials obtained by the traditional manual investigation screening method, and finally 15 materials in

Table 7. Defoliation-sensitive materials jointly screened by defoliation rate and PSRI.

No.	Material Name	Defoliation Rate (%)	Lint Percentage (%)	PSRI	No.	Material Name	Defoliation Rate (%)	Lint Percentage (%)	PSRI
YT039	Mutant4	88.03	78.48	0.1795	YT100	Yumian 1	90.48	84.88	0.1852
YT047	Xinluzao 11	87.41	91.43	0.1720	YT101	Xinluzhong 68	92.15	87.04	0.1723
YT068	Han 241	85.81	86.32	0.1690	YT102	Xinluzhong 75	86.52	88.00	0.1733
YT074	Liaomian 9	85.95	97.08	0.1984	YT104	J206-5	88.56	91.86	0.1756
YT076	Shaan 416	87.43	76.67	0.1859	YT107	Guanmian 614	88.31	94.20	0.1607
YT078	Simian 2	95.12	82.72	0.1649	YT113	Guanmian V5	90.97	88.06	0.1694
YT083	Xinluzao 10	93.43	93.24	0.1852	YT125	Xinluzao 73	94.23	98.04	0.1609
YT099	Yuan 247-31	92.57	94.20	0.1819

were determined. The defoliation rates of these 15 cotton materials range from 85.81% to 95.12%, the cotton boll cracking rates range from 76.67% to 98.04%, and the PSRI values range from 0.1607 to 0.1984.

Among them, seven materials showed excellent defoliation effects (>85%) and high cotton boll cracking rates (>90%), specifically including Liaomian 9, Xinluzao 10, Yuan 247-31, J206-5, Xinluzao 11, Guomian 614, and Xinluzao 73. These materials can be used as valuable resources for the subsequent genetic improvement and breeding of cotton varieties. The research results indicate that using unmanned aerial vehicles equipped with vegetation indices for screening shows a high consistency with the manual investigation screening method in screening excellent defoliation materials. This proves that it is feasible to screen cotton breeding materials with excellent defoliation effects using unmanned aerial vehicle multispectral technology.

4. Discussion

4.1. Defoliation Effect Evaluation

The main cotton-producing areas in Xinjiang generally adopt mechanized harvesting, but the residue of leaves will increase the impurity rate of seed cotton and reduce the fiber quality. Cotton varieties sensitive to defoliants have a good defoliation effect, with rapid leaf shedding, promoting boll opening, and reducing mechanical harvesting impurities. The defoliation effect is crucial to the mechanical harvesting efficiency and cotton quality. Therefore, it is necessary to screen breeding materials sensitive to defoliants. In this study, different cotton germplasm resources were treated with defoliants and control experiments. The results showed that defoliants can effectively promote cotton leaf shedding and boll opening, significantly improving the defoliation and boll opening speed of cotton. This indicates that defoliants can effectively promote cotton leaf shedding and boll opening, which is consistent with the research results of Chen et al. [29] and Gao et al. [30].

4.2. Comparison Between Traditional Survey Methods and UAV Multispectral Technology

On the traditional investigation method, the evaluation of cotton defoliation effect relies on manual investigation, including counting the number of cotton plant leaves and bolls before and after pesticide application. This method is time-consuming and error-prone. Moreover, when evaluating large cotton fields, due to vision limitations and subjective factors, it is difficult to accurately assess. The combination of high spatial resolution remote sensing and intelligent agricultural machinery provides a strong technical support for the acquisition of precise cultivated land information [31]. The method of using unmanned aerial vehicle (UAV) remote sensing as a data acquisition platform for crop monitoring can combine other auxiliary information. Previous studies have used UAV-related technologies to obtain relevant information such as crop water content [32], crop nitrogen content [33], radiation use efficiency [34], crop height information, and crop chlorophyll content. Currently, the monitoring research related to green leaves based on UAV is mainly focused on aspects such as leaf area index, biomass, and vegetation coverage. This study integrates UAV multispectral technology and vegetation index to evaluate the defoliation sensitivity of cotton, effectively overcoming the shortcomings of the traditional screening method, and providing a reference for the application of UAV multispectral technology in the field of cotton defoliation effect evaluation research.By quantitatively describing the growth indicators of crops, we can fully understand the growth status of cotton, providing a scientific basis for the development of precision agriculture.With the development of digital photography technology, obtaining crop growth information using digital images has become a more convenient and efficient method [35,36].

4.3. Discussion on the Application of Multi-Spectral Bands and Vegetation Indices

The study found that with the shedding of cotton leaves, the values of the red, green, and blue bands show an upward trend, while the values of the near-infrared (nir) and red-edge (red_edge) bands show a downward trend. When extracting spectral data in the field experiment, it was found that the reflectance of the green light band shows an upward trend. We guess that the possible reason for this phenomenon is that when cotton is in the boll opening stage, the water content of the leaves decreases and begins to fall off, which will lead to an increase in the reflectance of the green light by the leaves; and the reduction of water makes the medium that scatters the green light inside the leaves decrease, and more green light can be reflected. From the spectral curve, the reflection peak of the green light band may become more obvious. However, in addition to this physiological change, environmental factors may also be the cause of this phenomenon. When extracting data, we use the average value of all pixels within the selected range, and do not divide the cotton leaves or cotton bolls, etc., so this may lead to the change of this phenomenon. Therefore, in future experiments, the experiment can be further improved and perfected: 1. Using multi-scale spectroscopy for analysis, hyperspectral imaging technology can be added to analyze the spectral characteristics of different tissues in more detail; 2. Optimization of data acquisition methods, using image segmentation techniques or extracting texture features, such as through deep learning, to achieve precise distinction of leaves, cotton bolls [37], and background soil in UAV images; 3. Verifying through multiple environments to improve the application of UAV technology in the agricultural field.

Vegetation index is calculated through the reflectance of specific bands. It helps to reduce the interference of external factors such as soil and climate, thereby improving the accuracy and sensitivity of the target parameters. The shedding of cotton leaves is usually closely related to indicators such as chlorophyll content, leaf water status, and leaf area index (LAI); during the process of leaf shedding, chlorophyll degrades, and the photosynthetic capacity weakens, resulting in a decrease in the reflectance of the near-infrared band and an increase in the reflectance of the red light band. The shedding of cotton leaves leads to a reduction in the number of green leaves in the canopy. Researchers such as Yi [38] used remote sensing technology to monitor this change. For example, Li [39] diagnosed the premature senescence degree of cotton through different color spaces and eigenvalues, and achieved a good fitting effect. Yan [14] and the team used a quad-rotor unmanned aerial vehicle to collect multispectral images of cotton before and after spraying, combined with the maximum entropy and vegetation index threshold method to extract cotton leaf information, used the support vector machine to classify the cotton after spraying, and constructed a model through the comparison of field data to verify the application potential of unmanned aerial vehicle remote sensing technology in monitoring cotton defoliation. Previous studies have also confirmed the accuracy of the vegetation index as an indicator for monitoring changes in the cotton canopy. This study found that there is a significant positive correlation between PSRI and SIPI in the vegetation index and the defoliation rate, with the correlation coefficients exceeding 0.50; while MSRI, RVI, NDVI, SAVI, TVI, MVI, MCARI, EVI, and GNDVI show a significant negative correlation with the defoliation rate; indicating that the vegetation index can indicate the situation of cotton defoliation by reflecting the changes in the vegetation canopy.

4.4. Outlook

Looking forward to the future, with the continuous progress of unmanned aerial vehicle (UAV) remote sensing technology, data analysis methods, and artificial intelligence algorithms, the application of UAV technology in the field of precision agriculture is expected to provide more efficient solutions and promote agricultural production towards refined and intelligent management. Compared with the traditional manual observation and sampling methods, UAV remote sensing technology can achieve large-scale, real-time, and non-contact monitoring of the growth status of cotton plants. Traditional methods may be limited by the insufficient representativeness of samples and human errors, while UAV multispectral data provides a high spatial resolution and high-frequency data acquisition capability. Using this technology can effectively make up for the deficiencies of traditional investigation methods. UAV-related technologies provide a powerful technical support for screening sensitive materials suitable for cotton defoliation. To verify the stability and reliability of these screened sensitive materials, multi-point field experiments should be conducted in future studies to further evaluate their defoliation effect and boll opening rate performance. Meanwhile, continue to exploration of more vegetation indices and remote sensing technologies should be continued to improve a more accurate and efficient screening method for cotton defoliation-sensitive materials; in the future, it can cooperate with international agricultural research institutions to establish a global cotton defoliation sensitivity database using multispectral technology and promote the construction of a standardized evaluation system. This not only helps to promote the progress of cotton genetic improvement and breeding work, but also provides a strong technical support for the sustainable development of cotton production.

5. Conclusions

The test results show that the defoliant can significantly increase the defoliation rate and boll opening rate of cotton. After spraying the defoliant, there are significant differences in the leaf spectral characteristics of different cotton varieties. Through clustering analysis using the vegetation index PSRI, 24 materials with significant defoliation effects were screened out. A total of 37 defoliation-sensitive materials were obtained through the traditional manual investigation and screening method. A comparison of germplasm resources screened by these two methods was conducted. Finally, 15 breeding materials with excellent defoliation effects were determined. The defoliation rate and boll opening rate were between 85.81–95.12% and 76.67–98.04% respectively, and the PSRI value was within the range of 0.1607–0.1984, which verifies the effectiveness of the unmanned aerial vehicle multispectral technology in screening cotton breeding materials sensitive to defoliation.

Among the fifteen germplasm resources screened out, seven materials have a defoliation rate of over 85.00% and a boll opening rate of over 90%, including Liaomian 9 and Xinluzao 10, etc., showing good defoliation and boll opening characteristics, and are suitable as the preferred materials for germplasm resource improvement. The cracking rate of cotton bolls can be used as a relevant reference for yield prediction and harvest. Therefore, in subsequent experiments, we can continue to explore the potential changes in yield and quality of the materials with significant defoliation effects; by evaluating the cracking situation of these materials’ cotton bolls, we can further understand their practical production application value in cultivating and improving high-quality varieties.

In future studies, multi-point ecological zone verification experiments can be carried out, including cotton planting data under different geographical, climatic and soil conditions, to improve the accuracy of UAV multispectral technology. In addition, subsequent experiments can integrate texture features and temporal analysis in the multispectral data processing algorithm, which is expected to further improve the application of UAV technology in agriculture; the relevant models of boll opening rate and yield and quality can be further analyzed and constructed, thereby helping to select excellent breeding materials.

The results indicate that the use of drones equipped with the vegetation index for screening shows high consistency with manual survey methods in identifying superior defoliation materials. This confirms that using drone-based multispectral technology to select cotton breeding materials with excellent defoliation effects is a feasible approach. The drone multispectral screening technology based on the vegetation index effectively compensates for the limitations of traditional survey methods in investigating and selecting cotton defoliation, offering a more convenient and faster way to assess cotton defoliation status. This technology provides a more scientific and streamlined method for field management and germplasm selection. With high application value, it can offer technical support to agricultural producers for optimizing breeding selection and data surveys.