As core components in mechanical manufacturing, sliders have higher processing accuracy requirements than general parts and need their dimensional qualification to be strictly checked. Traditional manual measurement features low efficiency, high cost, and difficulty in meeting mass production needs. The introduction of machine vision technology is significant in realizing efficient, non-contact measurement of slider dimensions, avoiding damage, reducing costs, and promoting the intelligent development of the manufacturing industry. However, limitations in slider processing and camera imaging quality lead to surface disturbances such as scratches, rust spots, and edge blurring, making it difficult to achieve accurate dimensional measurements at limited resolution. To this end, this paper proposes a visual measurement method for slider dimensions combining sub-pixel counting and line segment cluster processing strategy: optimizing image quality through preprocessing to suppress interference factors; introducing a sub-pixel counting method to improve the Hough Transform algorithm for enhancing hole radius measurement accuracy; putting forward a line segment cluster processing strategy to accurately fit the target edge lines, compute the form and position dimensions of the slider, and design software for slider dimension measurement to visualize the measurement process. The experimental results show that the MAE, MADE, and RMSE of the method for slider sample measurement are 25 mu m, 0.174 %, and 35 mu m, respectively, and the ST is 0.909 s, which meet the demands for measurement accuracy and immediacy of slider dimensions. The research content of this paper provides important theoretical support for the automated dimensional inspection of production lines for slider-type parts.

The quality of forage point cloud data provides an important foundation for precision pasture management. Due to its high accuracy and real-time performance, LiDAR technology has become a potential tool for achieving fine management of forage. However, most existing mainstream point cloud processing algorithms are designed for structured or urban scenes. In natural pasture environments, where grass stems are dense, postures are complex, the ground is irregular, and noise points are widespread, traditional methods show significant shortcomings in registration, denoising, and segmentation, especially in ground extraction and grass segmentation. Therefore, a more targeted and robust preprocessing scheme is urgently needed. In this study, a dedicated point cloud preprocessing method for forage scenarios is proposed. First, an improved Weighted ICP (W-ICP) algorithm is used for data registration to obtain a complete and coherent point cloud model. Then, noise is removed by combining multi-level filtering and deep learning-based denoising. Next, an improved DBSCAN algorithm is employed to segment the data and extract the forage regions. Finally, feature extraction and classification are carried out by introducing an attention mechanism and a feature enhancement module into the PointNet++ model. These improvements are not simple combinations, but organic integrations and adaptive adjustments among modules based on the characteristics of forage point clouds, aiming to improve the overall preprocessing quality and efficiency. The proposed method has been validated on point cloud datasets acquired in complex pasture environments. Experimental results show that the proposed CIM method achieves an accuracy of 97.0

To address the problems of uneven stubble height and high missed-cutting rate caused by the insufficient profiling capability of traditional forage harvesters in complex hilly terrain, this paper designs a three-degrees-of-freedom (DOF) profiling header primarily for typical hilly terrain with gentle slopes of 8-15 degrees. Through pitch, roll, and height adjustments, it stably maintains stubble height at 150 mm. Subsequently, geometric analysis and structural optimization achieved kinematic decoupling among all degrees of freedom, thereby overcoming the inherent limitations of the two-DOF header, such as poor adaptability to longitudinal slope and strong adjustment coupling. Three-dimensional modeling was completed in SolidWorks, multibody dynamics simulation was performed in ADAMS, and a profiling control system incorporating a hydraulic system, multi-source sensor fusion, and a fuzzy PID controller was built. The dynamics simulation results show that under the working conditions of 15 degrees longitudinal and 10 degrees transverse slopes, the stubble height error of the header is controlled within 10%, the attitude angle adjustment error is less than 0.5 degrees, and the dynamic response is excellent. Prototype field tests showed that, compared with the two-DOF header, the three-DOF profiling header improved the stubble height stability by about 35%, reduced the missed-cutting rate by about 5%, and increased the operating efficiency by about 15%. No cutting blade contact with the soil occurred, verifying the rationality of the mechanism design and its adaptability to terrain. This study provides an effective technical solution for improving the mechanization level of forage harvesting in hilly and mountainous areas.

To address the challenges of accurately extracting features from elongated scratches, irregular defects, and small-scale surface flaws on high-precision linear guide rails, this paper proposes a novel instance segmentation algorithm tailored for guide rail surface defect detection. The algorithm integrates the YOLOv8 instance segmentation framework with deformable convolutional networks and multi-scale feature fusion to enhance defect feature extraction and segmentation performance. A dedicated guide rail surface Defect (GSD) segmentation dataset is constructed to support model training and evaluation. In the backbone, the DCNv3 module is incorporated to strengthen the extraction of elongated and irregular defect features while simultaneously reducing model parameters. In the feature fusion network, a multi-scale feature fusion module and a triple-feature encoding module are introduced to jointly capture global contextual information and preserve fine-grained local defect details. Furthermore, a Channel and Position Attention Module (CPAM) is employed to integrate global and local features, improving the model’s sensitivity to channel and positional cues of small-target defects and thereby enhancing segmentation accuracy. Experimental results show that, compared with the original YOLOv8n-Seg, the proposed method achieves improvements of 3.9% and 3.8% in Box and Mask mAP50, while maintaining a real-time inference speed of 148 FPS. Additional evaluations on the public MSD dataset further demonstrate the model’s strong versatility and robustness.

Accurate measurement of plant height in leafy vegetables is challenging due to their short stature, high planting density, and severe canopy occlusion during later growth stages. These factors often limit the reliability of single-plant monitoring across the full growth cycle in open-field environments. To address this, we propose a multi-temporal point cloud alignment method for accurate plant height measurement, focusing on Choy Sum (Brassica rapa var. parachinensis). The method estimates plant height by calculating the vertical distance between the canopy and the ground. Multi-temporal point cloud maps are reconstructed using an enhanced Oriented FAST and Rotated BRIEF-Simultaneous Localization and Mapping (ORB-SLAM3) algorithm. A fixed checkerboard calibration board, leveled using a spirit level, ensures proper vertical alignment of the Z-axis and unifies coordinate systems across growth stages. Ground and plant points are separated using the Excess Green (ExG) index. During early growth stages, when the soil is minimally occluded, ground point clouds are extracted and used to construct a high-precision reference ground model through Cloth Simulation Filtering (CSF) and Kriging interpolation, compensating for canopy occlusion and noise. In later growth stages, plant point cloud data are spatially aligned with this reconstructed ground surface. Individual plants are identified using an improved Euclidean clustering algorithm, and consistent measurement regions are defined. Within each region, a ground plane is fitted using the Random Sample Consensus (RANSAC) algorithm to ensure alignment with the X-Y plane. Plant height is then determined by the elevation difference between the canopy and the interpolated ground surface. Experimental results show mean absolute errors (MAEs) of 7.19 mm and 18.45 mm for early and late growth stages, respectively, with coefficients of determination (R2) exceeding 0.85. These findings demonstrate that the proposed method provides reliable and continuous plant height monitoring across the full growth cycle, offering a robust solution for high-throughput phenotyping of leafy vegetables in field environments.

Convolutional neural networks (CNNs) continue to present challenges in the form of redundant network architectures, high computational intensity, and difficulties in deployment on an expanding array of embedded devices for the practical task of multi-country currency image recognition. This paper proposes a new lightweight multinational banknote recognition model, designated CA-DSC-RepVGG. The backbone network employs the RepVGG-A0 network, and a lightweight CA coordinate attention mechanism is introduced after the residual structure of the backbone network. This enhances the feature extraction capability by emphasizing the information representation. Concurrently, the conventional convolution in the residual structure is enhanced through the utilization of depth-separable convolution. Subsequently, the CA-DSC-RepVGG model is deployed to the rk3568 embedded device. A series of comparative experiments were conducted on three contemporary banknote image datasets: Australian Dollar, Euro, and US Dollar. The experimental results demonstrate that CA-DSC-RepVGG enhances accuracy by 1.05

Accurately and precisely obtaining field crop information is crucial for evaluating the effectiveness of rice transplanter operations. However, the working environment of rice transplanters in paddy fields is complex, and data obtained solely from GPS devices installed on agricultural machinery cannot directly reflect the specific information of seedlings, making it difficult to accurately evaluate the quality of rice transplanter operations. This study proposes a CAD-UNet model for detecting rice seedling rows based on low altitude orthorectified remote sensing images, and uses evaluation indicators such as straightness and parallelism of seedling rows to evaluate the operation quality of the rice transplanter. We have introduced convolutional block attention module (CBAM) and attention gate (AG) modules on the basis of the original UNet network, which can merge multiple feature maps or information flows together, helping the model better select key areas or features of seedling rows in the image, thereby improving the understanding of image content and task execution performance. In addition, in response to the characteristics of dense and diverse shapes of seedling rows, this study attempts to integrate deformable convolutional network version 2 (DCNv2) into the UNet network, replacing the original standard square convolution, making the sampling receptive field closer to the shape of the seedling rows and more suitable for capturing various shapes and scales of seedling row features, further improving the performance and generalization ability of the model. Different semantic segmentation models are trained and tested using low altitude high-resolution images of drones, and compared. The experimental results indicate that CAD-UNet provides excellent results, with precision, recall, and F1-score reaching 91.14%, 87.96%, and 89.52%, respectively, all of which are superior to other models. The evaluation results of the rice transplanter’s operation effectiveness show that the minimum and maximum straightnessof each seedling row are 4.62 and 13.66 cm, respectively, and the minimum and maximum parallelismbetween adjacent seedling rows are 5.16 and 23.34 cm, respectively. These indicators directly reflect the distribution of rice seedlings in the field, proving that the proposed method can quantitatively evaluate the field operation quality of the transplanter. The method proposed in this study can be applied to decision-making models for farmland crop management, which can help improve the efficiency and sustainability of agricultural operations.

Choy sum (Brassica rapa var. parachinensis) is a commonly grown leafy vegetable, primarily harvested for its stem. Accurate stem segmentation is crucial for accurate harvesting, yet the visual similarity between choy sum stems and leaves poses challenges for traditional supervised learning methods, making data labeling costly and affecting segmentation accuracy. This study introduces AD-DMT, an enhanced Dynamic Mutual Training (DMT) algorithm for semi-supervised segmentation, which improves on the original framework by incorporating: 1) Introduction of data augmentation strategies such as CutMix, brightness, and contrast adjustments to alleviate model generalization difficulties caused by data homogeneity; 2) The design of adaptive loss weights re-scaled factor (71 and 72) dynamically adjusts the balance between mutual learning and entropy minimization based on training epochs; 3) A dynamic temperature coefficient is incorporated to enhance divergent learning in training by modulating Softmax output. For validation, images of field-grown choy sum were captured to evaluate AD-DMT's performance under different labeled data ratios (1/2, 1/4, 1/8, 1/20). The results demonstrate efficient segmentation across all conditions, with mIoU values exceeding 84.0 %. Notably, even with minimal labeled data (1/20 ratio), AD-DMT achieved a 4.04 % improvement in mIoU over the baseline. Building on these segmentation results, we further determined the optimal cutting points of choy sum stems by using skeleton extraction and corner detection algorithms, calculating the three-dimensional coordinates of these points with depth images, achieving an average vertical offset error (VOE) within 6.29 mm.

The existing forage harvester header cannot automatically adjust the height and inclination during operation, resulting in uneven stubble height of forage, which, in turn, affects the efficiency of harvesting and the quality of forage regeneration. To address this issue, this study conducted the design and experimentation of a 2-degrees-of-freedom (DOF) profiling header. Firstly, this study designed an adaptive profiling header with 2-DOF adjustment, which was realized by the height adjustment mechanism and the tilt angle adjustment mechanism. The relationship model between the profiling device and the attitude of the header was established so that the header can acquire ground undulation in real time through the angle sensor of the profiling device. In order to verify the rationality of the header design, a co-simulation model of ADAMS and MATLAB/Simulink was built, and the header attitude control system was designed based on the fuzzy PID algorithm. The co-simulation results show that the header height (H) is always kept around 150 mm during the forwarding process of the harvester, with a maximum error of 5.8 mm, and the average relative error (REH) and root mean square error (RMSEH) were 1.4% and 2.6 mm, respectively, and the maximum error of the tilt angle (γ) of the header is 0.53° and the RMSEγ is 0.22°, which indicates that the header profiling mechanism can accurately reflect the undulation of the terrain and the header attitude control system has good robustness. Finally, the test platform was built and tested in a grassland. The test results show that the average height of the header is 149.8 mm, the maximum error is 7.5 mm, and the REH and RMSEH are 3.4% and 5.3 mm, respectively. The average error of the header inclination is 0.34°, and the maximum error is 0.57°. The test results indicate that the header can realize the adaptive adjustment of height and inclination, and the control system has high precision, stability and reliability, meeting the demand of automatic regulation of header attitude of a forage harvester.

Soil fertility is a critical factor in agricultural production, directly impacting crop growth, yield, and quality. To achieve precise agricultural management, accurate spatial interpolation of soil parameters is essential. This study developed a new interpolation prediction framework that combines Radial Basis Function (RBF) neural networks with Inverse Distance Weighting (IDW), termed the IDW-RBFNN. This framework initially uses the IDW method to apply preliminary weights based on distance to the data points, which are then used as input for the RBF neural network to form a training dataset. Subsequently, the RBF neural network further trains on these data to refine the interpolation results, achieving more precise spatial data interpolation. We compared the interpolation prediction accuracy of the IDW-RBFNN framework with ordinary Kriging (OK) and RBF methods under three different parameter settings. Ultimately, the IDW-RBFNN demonstrated lower error rates in terms of RMSE and MRE compared to direct RBF interpolation methods when adjusting settings based on different power values, even with a fixed number of data samples. As the sample size decreases, the interpolation accuracy of OK and RBF methods is significantly affected, while the error of IDW-RBFNN remains relatively low. Considering both interpolation accuracy and resource limitations, we recommend using the IDW-RBFNN method (p = 2) with at least 60 samples as the minimum sampling density to ensure high interpolation accuracy under resource constraints. Our method overcomes limitations of existing approaches that use fixed steady-state distance decay parameters, providing an effective tool for soil fertility monitoring in delta regions.

The rapid and accurate identification of sugarcane internodes is of great significance for tasks such as field operations and precision management in the sugarcane industry, and it is also a fundamental task for the intelligence of the sugarcane industry. However, in complex field environments, traditional image processing techniques have low accuracy, efficiency, and are mainly limited to server-side processing. Meanwhile, the sugarcane industry requires a large amount of manual involvement, leading to high labor costs. In response to the aforementioned issues, this paper employed YOLOv5s as the original model algorithm, incorporated the K-means clustering algorithm, and added the CBAM attention module and VarifocalNet mechanism to the algorithm. The improved model is referred to as YOLOv5s-KCV. We implemented the YOLOv5s-KCV algorithm on Jetson TX2 edge computing devices with a well-configured runtime environment, completing the design and development of a real-time sugarcane internode identification system. Through ablation experiments, comparative experiments of various mainstream visual recognition network models, and performance experiments conducted in the field, the effectiveness of the proposed improvement method and the developed real-time sugarcane internode identification system were verified. The experimental results demonstrate that the improvement method of YOLOv5s-KCV is effective, with an algorithm recognition accuracy of 89.89%, a recall rate of 89.95%, and mAP value of 92.16%, which respectively increased by 6.66%, 5.92%, and 7.44% compared to YOLOv5s. The system underwent performance testing in various weather conditions and at different times in the field, achieving a minimum recognition accuracy of sugarcane internodes of 93.5%. Therefore, the developed system in this paper can achieve real-time and accurate identification of sugarcane internodes in field environments, providing new insights for related work in sugarcane field industries.

The rapid and accurate detection of Chinese flowering cabbage ripeness and the counting of Chinese flowering cabbage are fundamental for timely harvesting, yield prediction, and field management. The complexity of the existing model structures somewhat hinders the application of recognition models in harvesting machines. Therefore, this paper proposes the lightweight Cabbage-YOLO model. First, the YOLOv8-n feature pyramid structure is adjusted to effectively utilize the target’s spatial structure information as well as compress the model in size. Second, the RVB-EMA module is introduced as a necking optimization mechanism to mitigate the interference of shallow noise in the high-resolution sounding layer and at the same time to reduce the number of parameters in this model. In addition, the head uses an independently designed lightweight PCDetect detection head, which enhances the computational efficiency of the model. Subsequently, the neck utilizes a lightweight DySample upsampling operator to capture and preserve underlying semantic information. Finally, the attention mechanism SimAm is inserted before SPPF for an enhanced ability to capture foreground features. The improved Cabbage-YOLO is integrated with the Byte Tracker to track and count Chinese flowering cabbage in video sequences. The average detection accuracy of Cabbage-YOLO can reach 86.4%. Compared with the original model YOLOv8-n, its FLOPs, the its number of parameters, and the size of its weights are decreased by about 35.9%, 47.2%, and 45.2%, respectively, and its average detection precision is improved by 1.9% with an FPS of 107.8. In addition, the integrated Cabbage-YOLO with the Byte Tracker can also effectively track and count the detected objects. The Cabbage-YOLO model boasts higher accuracy, smaller size, and a clear advantage in lightweight deployment. Overall, the improved lightweight model can provide effective technical support for promoting intelligent management and harvesting decisions of Chinese flowering cabbage.

Accurately detecting the maturity and 3D position of flowering Chinese cabbage (Brassica rapa var. chinensis) in natural environments is vital for autonomous robot harvesting in unstructured farms. The challenge lies in dense planting, small flower buds, similar colors and occlusions. This study proposes a YOLOv8-Improved network integrated with the ByteTrack tracking algorithm to achieve multi-object detection and 3D positioning of flowering Chinese cabbage plants in fields. In this study, C2F-MLCA is created by adding a lightweight Mixed Local Channel Attention (MLCA) with spatial awareness capability to the C2F module of YOLOv8, which improves the extraction of spatial feature information in the backbone network. In addition, a P2 detection layer is added to the neck network, and BiFPN is used instead of PAN to enhance multi-scale feature fusion and small target detection. Wise-IoU in combination with Inner-IoU is adopted as a new loss function to optimize the network for different quality samples and different size bounding boxes. Lastly, ByteTrack is integrated for video tracking, and RGB-D camera depth data are used to estimate cabbage positions. The experimental results show that YOLOv8-Improve achieves a precision (P) of 86.5% and a recall (R) of 86.0% in detecting the maturity of flowering Chinese cabbage. Among them, mAP50 and mAP75 reach 91.8% and 61.6%, respectively, representing an improvement of 2.9% and 4.7% over the original network. Additionally, the number of parameters is reduced by 25.43%. In summary, the improved YOLOv8 algorithm demonstrates high robustness and real-time detection performance, thereby providing strong technical support for automated harvesting management.

To verify the dynamic model of a paddy-field laser grader and solve the problem of the unstable high-speed operation of the grader, a method for measuring the absolute pose parameters of a moving rigid body using a monocular camera is proposed. The proposed method does not require calibration beforehand. Using more than six non-coplanar control points symmetrically arranged in the rigid-body and world coordinate systems, the matrices of rotation and translation between the camera and two coordinate systems are obtained and the absolute pose of the rigid body is measured. In this paper, formulas of the absolute pose measurement of a moving rigid body are deduced systematically and the complete implementation is presented. Position and attitude measurement experiments carried out on a three-axis precision turntable show that the average absolute error in the attitude angle of a moving rigid body measured by an uncalibrated camera at different positions changes by no more than 0.2 degrees. An analysis of the three-dimensional coordinate errors of the centroid of a moving rigid body shows little deviation in measurements taken at the three camera positions, with the maximum deviation of the average absolute error being 0.53 cm and the maximum deviation of the standard deviation being 0.66 cm. The proposed method can measure the absolute pose of a rigid body and is insensitive to the position of the camera in the measurement process. This work thus provides guidance for the repeated measurement of the absolute pose of a moving rigid body using a monocular camera.

Objects in agricultural soils will seriously affect the farming operations of agricultural machinery. At present, it still relies on human experience to judge abnormal Gounrd-penetrting Radar (GPR) signals. It is difficult for traditional image processing technology to form a general positioning method for the randomness and diversity characteristics of GPR signals in soil. Although many scholars had researched a variety of image-processing techniques, most methods lack robustness. In this study, the deep learning algorithm Mask Region-based Convolutional Neural Network (Mask-RCNN) and a geometric model were combined to improve the GPR positioning accuracy. First, a soil stratification experiment was set to classify the physical parameters of the soil and study the attenuation law of electromagnetic waves. Secondly, a SOIL-GPR geometric model was proposed, which can be combined with Mask-RCNN's MASK geometric size to predict object sizes. The results proved the effectiveness and accuracy of the model for position detection and evaluation of objects in soils; then, the improved Mask RCNN method was used to compare the feature extraction accuracy of U-Net and Fully Convolutional Networks (FCN); Finally, the operating speed of agricultural machinery was simulated and designed the A-B survey line experiment. The detection accuracy was evaluated by several indicators, such as the survey line direction, soil depth false alarm rate, Mean Average Precision (mAP), and Intersection over Union (IoU). The results showed that pixel-level segmentation and positioning based on Mask RCNN can improve the accuracy of the position detection of objects in agricultural soil effectively, and the average error of depth prediction is 2.87 cm. The results showed that the detection technology proposed in this study integrates the advantage of soil environmental parameters, geometric models, and artificial intelligence algorithms to provide a high-precision and technical solution for the GPR non-destructive detection of soils.

Broken eggs can be harmful to human health but are also unfavorable for transportation and production. This study proposes a video-based detection model for the real-time detection of broken eggs regarding unwashed eggs in dynamic scenes. A system capable of the continuous rotation and translation of eggs was designed to display the entire surface of an egg. We added CA into the backbone network, fusing BiFPN and GSConv with the neck to improve YOLOv5. The improved YOLOV5 model uses intact and broken eggs for training. In order to accurately judge the category of eggs in the process of movement, ByteTrack was used to track the eggs and assign an ID to each egg. The detection results of the different frames of YOLOv5 in the video were associated by ID, and we used the method of five consecutive frames to determine the egg category. The experimental results show that, when compared to the original YOLOv5, the improved YOLOv5 model improves the precision of detecting broken eggs by 2.2%, recall by 4.4%, and mAP:0.5 by 4.1%. The experimental field results showed an accuracy of 96.4% when the improved YOLOv5 (combined with ByteTrack) was used for the video detection of broken eggs. The video-based model can detect eggs that are always in motion, which is more suitable for actual detection than a single image-based detection model. In addition, this study provides a reference for the research of video-based non-destructive testing.

Chinese flowering cabbage (Brassica campestris L. ssp. chinensis var. utilis Tsen et Lee) is an important leaf vegetable originating from southern China. Its planting area is expanding year by year. Accurately judging its maturity and determining the appropriate harvest time are crucial for production. The open state of Chinese flowering cabbage buds serves as a crucial maturity indicator. To address the challenge of accurately identifying Chinese flowering cabbage buds, we introduced improvements to the feature fusion approach of the YOLOv5 (You Only Look Once version 5) algorithm, resulting in an innovative algorithm with a dynamically adjustable detection head, named FPNDyH-YOLOv5 (Feature Pyramid Network with Dynamic Head-You Only Look Once version 5). Firstly, a P2 detection layer was added to enhance the model’s detection ability of small objects. Secondly, the spatial-aware attention mechanism from DyHead (Dynamic Head) for feature fusion was added, enabling the adaptive fusion of semantic information across different scales. Furthermore, a center-region counting method based on the Bytetrack object tracking algorithm was devised for real-time quantification of various categories. The experimental results demonstrate that the improved model achieved a mean average precision (mAP@0.5) of 93.9%, representing a 2.5% improvement compared to the baseline model. The average precision (AP) for buds at different maturity levels was 96.1%, 86.9%, and 98.7%, respectively. When applying the trained model in conjunction with Bytetrack for video detection, the average counting accuracy, relative to manual counting, was 88.5%, with class-specific accuracies of 90.4%, 80.0%, and 95.1%. In conclusion, this method facilitates relatively accurate classification and counting of Chinese flowering cabbage buds in natural environments.

Accurate spatial information of farmland in small farms is very important to provide operable information to farmers, managers and decision makers. However, small farms have small area, irregular shape, and use a variety of planting crops, which makes their boundaries blurred, and the standard edge detection algorithm cannot accurately segment the farmland boundary. Therefore, the automatic delimitation of fields in small farms is a challenging task. Aiming at the above problems, this paper proposes an example segmentation method of Mask R-CNN based on dual attention mechanism feature pyramid network (DAFPN) to describe small farms. DAFPN is composed of two attention modules: spatial attention module (SPA) and channel attention module (CHA) to enhance its feature extraction ability. Spatial attention module (SPA) generates spatial attention map by using the spatial relationship of features, and generates information to be emphasized or suppressed in spatial location; The channel attention module (CHA) learns an adaptive channel merging method based on the attention mechanism. Our proposed DAFPN can be easily inserted into the existing FPN model. We have conducted extensive experimental analysis on very high resolution (VHR) satellite images based on the Mask R-CNN deep learning framework of DAFPN. The standard COCO dataset evaluation index and F1-score evaluation strategy are used to compare the algorithm. AP50, AP75 and F1-score reach 82.86%, 55.51% and 70.90% respectively, which is 8.7%, 8.31% and 6.87% higher than Mask R-CNN respectively. Our results highlight the ability of Mask R-CNN based on DAFPN to accurately depict small farms in VHR satellite images, which lays a foundation for the automatic segmentation of small farms.

A full understanding of the growth and distribution of tree roots is conducive to guiding precision irrigation, fertilization, and other agricultural work during agricultural production. Detecting tree roots with a ground-penetrating radar is a repeatable detection method that does no harm to the earth surface and tree roots. In this research, a rapid and accurate automatic detection was conducted on hyperbolic waveforms formed by root targets in B-scan images based on YOLOv5s. Following this, the regions of interest containing target hyperbolas were generated. Three or more coordinate points on the hyperbola were selected according to the three-point fixed circle (TPFC) method to locate the root system and estimate the root diameter. The results show that the accuracy of hyperbola detection using YOLOv5s was 96.7%, the recall rate was 86.6%, and the detection time of a single image was only 13 ms. In the simulation image, the TPFC method was used to locate the root system and estimate the root diameter through three different frequency antennas (500 MHz, 750 MHz, and 1000 MHz). A more accurate result was obtained when the antenna frequency was 1000 MHz, with the average distance error of root system positioning being 3.17 cm, and the slope and R2 of the linear fitting result between the estimated root diameter and the actual one being 1.029 and 0.987, respectively. Verified by the pre-buried root test and wilderness field test, both root localization and root diameter estimation in our research were proved to gain good results and conform to the rules found in simulation experiments. Therefore, we believe that this method can quickly and accurately detect the root system, locate and estimate the root diameter, and provide a new perspective for the non-destructive detection of the root system and the three-dimensional reconstruction of the root system.