This study focused on three tea cultivars from the Jianghan Plain to construct an inversion model between multispectral features and chlorophyll content in tea leaves. Based on 120 samples across two growth stages, indoor multispectral imaging technology was used to simultaneously acquire leaf multispectral data and SPAD values. Through the analysis of the spectral-chlorophyll response mechanism and the evaluation of feature wavelength autocorrelation, the Iteratively Retained Informative Variables (IRIV) algorithm was integrated for feature selection. An evaluation system consisting of seven machine learning models, including Partial Least Squares Regression (PLSR) and Support Vector Regression (SVR), was established. The results showed that the model combining the adjacent band change rate features selected by IRIV with Multiple Linear Regression (MLR) achieved the optimal inversion accuracy (R2=0.785, RMSE=4.241). Additionally, the vegetation index-MLR combination (R2=0.791, RMSE=4.222) and the mixed feature-LASSO combination (R2=0.773, RMSE=4.403) performed prominently under different feature dimensions. This study provides a feature engineering scheme with strong interpretability and a model optimization path for hyperspectral nondestructive detection of tea physiological parameters.

To address the challenges of hyperspectral data redundancy, small-target segmentation difficulty, and insufficient model real-time performance in high-precision online detection of foreign matter and kernel breakage in machine-harvested soybeans, this study proposes a collaborative detection method based on "feature wavelength optimization + MobileNetV4-Unet-SGCPNet hybrid network". First, 18 key feature bands were screened from 400 - 1000 nm hyperspectral data using successive projection algorithm (SPA) and competitive adaptive reweighted sampling (CARS), constructing a multi-source feature spectral image dataset. Subsequently, a MobileNetV4-Unet-SGCPNet hybrid network was designed, with a lightweight MobileNetV4 as the encoder, combined with the symmetric encoder-decoder structure of Unet and the spatial detail-guided context propagation module (SGCP) to achieve high-precision segmentation of broken grains, complete grains, and impurities. Finally, pixel-wise voting was employed to fuse multi-band feature information, enhancing the model's generalization capability. The results demonstrate that: on the test set, the model achieves an average intersection over - union of 89.69 % for soybean component recognition, with a mean precision average of 94.55 %, a mean precision of 93.63 %, a frame rate of 4.77 FPS, a parameter count of only 2.86 MB, and a computational load of 35.78 GFLOPs. Compared to mainstream models, this method reduces parameters by 97.2 % and computational cost by 97.8 %, while the average intersection - over - union drops by only 6.2 %, with a frame rate improvement of over 5 times, striking a significant balance between detection accuracy and real-time performance. Crossvariety and cross-device validations further confirm that the model effectively adapts to morphological and spectral variations across different soybean varieties, exhibiting strong generalization ability. This study provides a core algorithmic foundation for online monitoring systems of intelligent harvester operation quality, offering critical support for enhancing the commercial value of machine-harvested soybeans and advancing the intelligence level of agricultural machinery.

To address the issue that mismatches between operating conditions and control parameters during combine harvester operations lead to increased grain breakage rate, loss rate, and impurity rate, this study proposes a reinforcement learning–based optimization control method for operational parameters, supported by self-developed grain quality and loss rate sensors. A reinforcement learning simulation environment for the threshing and cleaning system of the combine harvester is established. On this basis, the effects of four types of reward function components—optimal value, constant-offset optimal value, reference-offset optimal value, and minimum power consumption—as well as their combinations, on the control effectiveness of the reinforcement learning algorithm are systematically analyzed. Simulation results demonstrate that the combination of multiple reward functions improves the performance of the algorithm within the range of operating conditions evaluated in this study. The four-combined reward reduces impurity rate by 6.6

To achieve the rapid and accurate classification and identification of soybean components, this study selected soybeans harvested by the 4LZ-1.5 soybean combine harvester as the research subject. Hyperspectral images of soybean samples were collected using the Pika L spectrometer, and spectral information was extracted from the regions of interest (ROI) in the images. Eight preprocessing methods, including baseline correction (BC), moving average (MA), Savitzky–Golay derivative (SGD), normalization, standard normal variate transformation (SNV), multiplicative scatter correction (MSC), first derivative (DS), and Savitzky–Golay smoothing (SGS), were applied to the raw spectral data to eliminate irrelevant information. Feature wavelengths were selected using the successive projections algorithm (SPA) and the competitive adaptive reweighted sampling (CARS) algorithm to reduce spectral redundancy and enhance model detection performance, retaining eight and ten feature wavelengths, respectively. Subsequently, a random forest (RF) model was developed for soybean component classification. The model parameters were optimized using particle swarm optimization (PSO) and differential evolution (DE) algorithms to improve performance. Experimental results showed that the RF classification model based on SPA-BC preprocessed spectra and DE-tuned parameters achieved an optimal prediction accuracy of 1.0000 during training. This study demonstrates the feasibility of using hyperspectral imaging technology for the rapid and accurate detection of soybean components, providing technical support for the assessment of breakage and impurity levels during soybean harvesting and storage processes. It also offers a reference for the development of future machine-harvested soybean breakage and impurity detection systems.

Harvest quality is one of the key indicators for evaluating the performance of mechanized wheat harvesting. To enable real-time monitoring of harvest quality during the wheat combine harvesting process, this study integrates an online detection system specifically designed for wheat harvest quality. The system is capable of real-time monitoring of threshing and cleaning losses, breakage rate, and impurity rate. To assess the effectiveness of the online detection system, field experiments were conducted. The experimental results showed that the average threshing and cleaning loss rates of the harvester measured manually were 0.69% and 0.75%, respectively, while those detected by the online system were 0.52% and 0.50%, respectively. The average grain breakage rates obtained through manual measurement were 0.67% and 0.58%, whereas the system detected breakage rates of 0.49% and 0.52%, respectively. For impurity rate, manual measurements yielded average values of 0.81% and 0.69%, while the system recorded 0.73% and 0.67%, respectively. The results demonstrate that the developed online wheat harvest quality detection system can effectively perform real-time assessments of harvest quality during combine harvesting operations.

To address the challenges of insufficient accuracy and low efficiency in full-width harvesting during autonomous harvester navigation, this study proposes a lightweight image segmentation model based on an improved U-Net architecture with enhanced Swin-Transformer modules. The symmetric encoder-decoder network integrates three refined Swin-Transformer blocks in both pathways, coupled with skip connections to preserve spatial features during downsampling and upsampling. Trained and validated on a dataset of wheat field images captured by a ZED stereo camera, the model achieved state-of-the-art performance with 95.49 % mean Intersection over Union (MIoU), 98.38 % Mean Pixel Accuracy (MPA), and real-time processing at 55.5 frames per second (FPS), while maintaining exceptional computational efficiency (0.68 M parameters, 19.788 GFLOPs). Comparative experiments with DeeplabV3 + and conventional U-Net variants demonstrated superior segmentation accuracy and operational efficiency. Navigation line extraction using region overlapping and polygon midpoint fitting methods yielded an average deviation error of 6.46 pixels, confirming the model's practical applicability for precision full-width harvesting in agricultural automation.

To address the limitation of existing agricultural unmanned plant protection equipment in perceiving crop growth status in real time during the maize seedling stage, this study proposes a crop row extraction method based on image processing. A crop semantic segmentation network was developed using the UNet framework, with VGG19 as the encoder and transposed convolution as the decoder. Model testing demonstrated that the segmentation network achieved accuracy rates of 0.9865 on the training set and 0.9864 on the validation set, with corresponding loss values of 0.0254 and 0.0270. In continuous processing scenarios, the average time for semantic segmentation per image was 120 milliseconds, while crop row extraction required 23 milliseconds.

This article aims to improve the real-time monitoring accuracy of the loss rate for grain combine harvesters by optimizing the sensor-sensitive plate structure, thereby addressing the problem of low detection efficiency in existing equipment. Based on Kirchhoff’s thin plate theory, COMSOL 6.0 software was utilized to conduct modal analysis and single-grain impact tests on rectangular and circular sensing plates fabricated from three materials: stainless steel, aluminum alloy, and cupronickel. The circular stainless steel sensing plate was identified as the optimal structure, whose natural frequency and sensitivity significantly outperform those of traditional rectangular plates. By integrating a signal processing strategy based on FFT (Fast Fourier Transform) spectrum analysis (band-pass filtering: 1.0~3.0 kHz, voltage threshold: 3.5 V) and a high-level duration counting algorithm, the system effectively distinguishes between grains and impurities and resolves the counting errors caused by multi-grain impacts and secondary rebounds. Field experiments demonstrate that the developed sensor exhibits strong anti-interference ability and high measurement accuracy, providing reliable technical support for reducing harvesting losses.

Studying the effect of variable fertilization during the jointing stage on winter wheat production in the rice-wheat rotation area is critical for evaluating the application effect and economic benefits of variable fertilization technology. The variable fertilization experiment of wheat was carried out in Jiangsu province by using the self-developed fertilizer applicator. Three fertilization methods were used to conduct a comparative analysis of the fertilization amount, population structure, and yield of winter wheat during the jointing stage. On this basis, the economic feasibility of variable fertilization during the jointing stage was evaluated. The experimental results showed that the control accuracy of variable-rate fertilization with fertilization equipment was greater than 95%. After variable fertilization, the coefficient of variation of NDVI values in the winter wheat canopy spectral data remained between 0.076 and 0.125, and the Christensen uniformity coefficient remained between 0.901 and 0.940. Compared with the traditional empirical balance method for quantitative fertilization of plots, the real-time variable fertilization plot used 13.6 kg/hm2 less fertilizer during the jointing stage. The findings validate that implementing variable fertilization can help reduce nitrogen fertilizer input, improve nitrogen fertilizer utilization efficiency, reduce environmental pollution, and enhance the sustainability of agricultural production.

Visual navigation is widely used in intelligent combine harvesters, but the existing algorithms do not have sufficiently high accuracy of the visual navigation line recognition under different sunlight conditions. To address this problem, this article proposes a sunlight-robust DeepLabV3+-based navigation line extraction method for combine harvesters. The navigation lines are extracted by constructing a new dataset and predicting the boundaries of the areas that have been and have not been cut. To address the problem that DeeplabV3+ is not sufficient light in the DCNN part, improvement is proposed by incorporating the MobileNetV2 module. In image segmentation, the prediction time is 22.5 ms, and the mean intersection over union (FMIOU) is 0.79. After image segmentation, the navigation lines are drawn using the line segment detection algorithm for the harvester. The proposed method is compared with other mainstream networks, and the prediction results are compared using the line segment detection method. The results show that this method can more quickly identify the navigation lines under different conditions of sunlight with less labeled data than the improved U-Net and DeeplabV3+, which uses Xception as the backbone. Compared to the traditional method and the improved U-Net, this method achieves good results and improves the recognition speed by 27 and 9 ms, respectively.

Deep learning and machine vision technology are widely applied to detect the quality of mechanized soybean harvesting. A clean dataset is the foundation for constructing an online detection learning model for the quality of mechanized harvested soybeans. In pursuit of this objective, we established an image dataset for mechanized harvesting of soybeans. The photos were taken on October 9, 2018, at a soybean experimental field of Liangfeng Grain and Cotton Planting Professional Cooperative in Guanyi District, Liangshan, Shandong, China. The dataset contains 40 soybean images of different qualities. By scaling, rotating, flipping, filtering, and adding noise to enhance the data, we expanded the dataset to 800 frames. The dataset consists of three folders, which store images, label maps, and record files for partitioning the dataset into training, validation, and testing sets. In the initial stages, the author devised an online detection model for soybean crushing rate and impurity rate based on machine vision, and research outcomes affirm the efficacy of this dataset. The dataset can help researchers construct a quality prediction model for mechanized harvested soybeans using deep learning techniques.

Using hyperspectral imaging technology for rapid, non-destructive detection of soybean grain moisture content provides technical support for high-quality soybean harvesting. A total of 90 samples of soybean grains from different varieties were collected, with hyperspectral images acquired in the wavelength range of 900-1700 nm. The moisture content of each soybean grain sample was determined using the direct drying method as specified in GB 5009.3-2016. The samples were divided into a calibration set and a prediction set based on a 4:1 ratio using the sample partitioning method of Joint X-Y Distance. Eight preprocessing methods were applied to the raw spectral data, including baseline correction, moving average, Savitzky-Golay filtering, normalization, standard normal variate transformation, multiple scatter correction, first derivative, and deconvolution. Feature wavelengths were then extracted using the successive projections algorithm and the competitive adaptive reweighted sampling algorithm. Finally, a partial least squares regression model for predicting the moisture content of soybean grains was developed based on these feature wavelengths. The results show that the correlation coefficient and the root mean square error of the optimal model for the prediction set were 0.92 and 0.2371, respectively. The moisture spectrum inversion model can precisely and rapidly predict the moisture content of soybean grains non-destructively, thereby determining the timing of mechanical soybean harvesting and enhancing the quality of soybean harvesting, storage, and processing.

Abstract Aiming at the problem of serious damage in the process of soybean harvesting and production, this paper analyzes the soybean harvesting process, and determines the main crushing forms of soybeans: epidermal breakage, splitting into two flaps, and overall crushing. The influence of different operation links on soybean crushing during the harvesting process, such as header feeding auger, horizontal seed spiral conveyor, horizontal miscellaneous residue spiral conveyor, and threshing roller was studied. The influence of the operating parameters such as the forward speed of the harvester, threshing roller speed, and deflector angle on the soybean crushing form was studied experimentally. The results showed that, the degree of influence on the proportion of sepidermal breakage from high to low was the forward speed of the harvester, deflector angle and threshing roller speed. The degree of influence on the proportion of splitting into two flaps from high to low was the deflector angle, threshing roller speed and the forward speed of the harvester. The degree of influence on the proportion of the overall crushing from high to low was the forward speed of the harvester, threshing roller speed and deflector angle.The research results can provide a reference for the subsequent structural improvement of soybean harvester and optimization of operating parameters.

Investigations regarding the use of centrifugal variable-rate fertilizer spreaders and their impact on agriculture production has become increasingly important for determining the effects and economic benefits of precision agriculture variable-rate fertilizer technology applied in the rice-and-wheat rotation areas of the middle and lower reaches of the Yangtze River in China. This study examines a self-developed centrifugal variable-rate fertilizer spreader based on real-time growth information of rice and wheat focusing on rice variety Nanjing 9108 as its research object. Field tests of variable-rate fertilizer application (compound fertilizer) were conducted in the Dongtai rice planting base, Jiangsu Province. Different areas were set up for artificial fertilizer spreading, traditional empirical balanced spreading, and real-time variable-rate spreading. The fertilizer application rates, rice growths, and yields of these fertilizer spreading methods for three consecutive years were compared and analyzed. Furthermore, the operational effectiveness of centrifugal variable-rate spreading was evaluated via partial budget analysis. The experimental results showed that the spreading application rate for the real-time variable-rate fertilization area decreased year by year by approximately 47.64 kg/hm2, whereas the application rate per unit area was reduced by approximately 15.88 %, compared with those for the traditional empirical balanced fertilization area. After the application of topdressing tillering fertilizer, rice growth in the real-time variable-rate fertilization area was better than in the others, whereas the spatial difference in its population structure was the lowest. There were differences in rice yield among the experimental areas: the average yield of the variable-rate fertilization area was 8996.70 kg/hm2, which was 5.93 % higher than that of the traditional empirical balanced fertilization area. According to the operational effectiveness analysis of real-time variable-rate fertilizer application in rice production, its fertilizer cost was only 504.72 RMB/hm2, whereas its yield profit was 17993.40 RMB/hm2. Thus, the fertilizer cost was reduced by 95.28 RMB/hm2, whereas the yield profit was increased by 1007.62 RMB/hm2, compared with those of the traditional empirical balanced fertilizer application. To sum up, this investigation verified that the newly developed centrifugal variable-rate fertilizer spreader, using real-time variable-rate fertilizer, can save on fertilizer and reduce its application, improve the population structure and spatial difference in crop growth, help increase crop yield and degree of balance, and produce economic, ecological, and social benefits. This study can provide a theoretical reference for promoting the pilot application of variable-rate fertilization technology in the rice-and-wheat rotation areas of the middle and lower reaches of the Yangtze River.

针对传统大豆联合收获机破碎率在线检测方法以人工检测耗时耗力且受人为主观因素影响的问题,提出基于DeepLabV3+网络的机收大豆破碎率在线检测方法.利用大豆图像在线采集装置获取联合收获机实时收获的大豆图像,使用标注软件对图像进行标注,构建数据集.为进一步提高网络训练速度,在DeepLabV3+网络中主干特征提取网络选用轻量级卷积网络MobileNetV2替代网络Xception;在预测部分,采用加黑边裁剪拼接的方式,提高图像分割精度.试验结果表明:基于DeepLabV3+网络模型对测试集大豆样本图像中破碎籽粒识别的综合评价指标F1值为89.49％,完整籽粒识别的综合评价指标F1值为93.93％;建立破碎率量化模型,进行台架试验,采用本文提出大豆破碎率在线检测方法检测结果平均值与人工检测结果平均值相对误差0.36％;为大豆联合收获机作业质量在线检测提供参考.

Wheat, one of the most important food crops in the world, is usually harvested mechanically by combine harvesters. The impurity rate is one of the most important indicators of the quality of wheat obtained by mechanized harvesting. To realize the online detection of the impurity rate in the mechanized harvesting process of wheat, a vision system based on the DeepLabV3+ model of deep learning for identifying and segmenting wheat grains and impurities was designed in this study. The DeepLabV3+ model construction considered the four backbones of MobileNetV2, Xception-65, ResNet-50, and ResNet-101 for training. The optimal DeepLabV3+ model was determined through the accuracy rate, comprehensive evaluation index, and average intersection ratio. On this basis, an online detection method of measuring the wheat impurity rate in mechanized harvesting based on image information was constructed. The model realized the online detection of the wheat impurity rate. The test results showed that ResNet-50 had the best recognition and segmentation performance; the accuracy rate of grain identification was 86.86%; the comprehensive evaluation index was 83.63%; the intersection ratio was 0.7186; the accuracy rate of impurity identification was 89.91%; the comprehensive evaluation index was 87.18%; the intersection ratio was 0.7717; and the average intersection ratio was 0.7457. In terms of speed, ResNet-50 had a fast segmentation speed of 256 ms per image. Therefore, in this study, ResNet-50 was selected as the backbone network for DeepLabV3+ to carry out the identification and segmentation of mechanically harvested wheat grains and impurity components. Based on the manual inspection results, the maximum absolute error of the device impurity rate detection in the bench test was 0.2%, and the largest relative error was 17.34%; the maximum absolute error of the device impurity rate detection in the field test was 0.06%; and the largest relative error was 13.78%. This study provides a real-time method for impurity rate measurement in wheat mechanized harvesting.

The level of mechanized harvesting of wheat in China has reached over 97%, and the impurity rate is one of the important indicators of mechanized wheat harvesting. In order to realize the online detection of the impurity rate in the wheat mechanized harvesting process, an online detection method of the wheat machine harvesting impurity rate was proposed based on the improved U-Net model combined with attention. Based on the wheat sample images collected by machine, the Labelme was used to manually label the images, and the images were enhanced by random rotation, scaling, shearing, and horizontal mirroring to construct a basic image dataset; an improved U-Net model combined with attention was designed. The model was classified and identified, and the offline training of the model was implemented under the torch 1.2.0 deep learning framework; the optimal offline model was transplanted to the Nvidia jetson tx2 development kit, and a quantification model of impurity rate was designed based on image information, so as to realize wheat on-line detection of impurity content in mechanized harvesting. The experimental results showed that the comprehensive evaluation index F1 of the improved U-Net model combined with attention was 76.64% and 85.70%, respectively, which were 10.33 percentage points and 2.86 percentage points higher than that of the standard U-Net, and 10.22 percentage points and 11.62 percentage points higher than that of DeepLabV3, which was 18.40 percentage points and 14.67 percentage points higher than that of PSPNet. Quantitative analysis of the detection results of impurity rate showed that in the bench test and field test, the average online detection of impurity rate of the device was 1.69% and 1.48%, respectively, which was higher than the manual detection by 0.26 percentage points and 0.13 percentage points. Qualitative analysis of the test results of impurity rate showed that whether it was a bench test or a field test, the test results of the device and the labor were all less than 2%. It was judged that the operation performance of the combine harvester during the test process met the national standards, and the test results were consistent. Therefore, the online detection method of wheat impurity rate proposed can provide technical support for the online quality control of wheat combined harvesting operations.

In this study, an online detection system of soybean crushed rate and impurity rate based on DeepLabV3+model was constructed. Three feature extraction networks, namely the MobileNetV2, Xception65, and ResNet-50 models, were adopted to obtain the best DeepLabV3+model through test analysis. Two well-established semantic segmentation networks, the improved U-Net and PSPNet, are used for mechanically harvested soybean image recognition and segmentation, and their performances are compared with the DeepLabV3+ model's performance. The results show that, of all the models, the improved U-Net has the best segmentation performance, achieving a mean intersection over union (FMIOU) value of 0.8326. The segmentation performance of the DeepLabV3+ model using the MobileNetV2 is similar to that of the U-Net, achieving FMIOU of 0.8180. The DeepLabV3+ model using the MobileNetV2 has a fast segmentation speed of 168.6 ms per image. Taking manual detection results as a benchmark, the maximum absolute and relative errors of the impurity rate of the detection system based on the DeepLabV3+ model with the MobileNetV2 of mechanized soybean harvesting operation are 0.06% and 8.11%, respectively. The maximum absolute and relative errors of the crushed rate of the same system are 0.34% and 9.53%, respectively.

Aiming at the demand by unmanned farms for unmanned operation in the entire process of field management, an unmanned plant protection robot for field management was developed based on a platform comprising a traditional high-clearance spray rod sprayer, integrated unmanned driving technology, image recognition technology, intelligent control technology, and precision operation technology. According to the agricultural machinery operation mode, agricultural machinery path planning, linear path tracking, and header path tracking algorithms were developed. Based on the overall structure and working principle of the chassis, the robot control system, steering control system, and operation control system were set. Based on the YOLOv5 image recognition algorithm, the crop–weed recognition model was developed. After 6000 rounds of training, the accuracy, recall, and mean average precision of the model were 87.7%, 84.5%, and 79.3%, respectively. Finally, a field experiment was carried out with the unmanned plant protection robot equipped with a complete system. Results show that the average lateral error of the robot is 0.036 m, the maximum lateral error is 0.2 m, the average root mean square error is 0.053 m, the average velocity error is 0.034 m/s, and the average root mean square error of velocity is 0.045 m/s when the robot works in a straight line. In weeding operations, the area ratio of weedy zones to field is 25%, which saves 75% of the herbicide compared to that dispensed in full spraying mode. The unmanned plant protection robot designed in this study effectively achieves machinery’s autonomous operation, providing valuable insights for research in unmanned farming and autonomous agricultural machinery.