Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Shengyong Xu has a total of 94 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • GLP-Mamba: A Global–Local Perception Mamba Model for Road Extraction from Optical Remote Sensing Imagery

    by Sheng Sun, Yu Chen, Xu Yu, Jianquan Li, Shengyong Xu, Changcai Yang

    Road extraction from high-resolution remote sensing imagery remains an active research topic in the remote sensing and intelligent transportation communities. Despite recent advances, existing methods still face significant challenges, including fragmented road representations and inaccurate boundary delineation. This paper proposes a road extraction model, termed GLP-Mamba, built upon the RS-Mamba architecture. The proposed model introduces a Global–Local Perception (GLP) module and a Structural Consistency Enhancement (SCE) Block. GLP refines shallow feature representations before fusion, while SCE performs residual structural refinement during decoding. Together, they improve semantic discrimination, boundary representation and structural continuity. Experiments are conducted on three road extraction datasets: The DeepGlobe Roads Dataset, the Massachusetts Roads Dataset, and the Jilin-1 Satellite Dataset. The results show that GLP-Mamba achieves the highest F1-score and Intersection over Union (IoU) among the compared methods. Ablation studies further verify the effectiveness of each component. Specifically, the OSS Block provides long-range contextual modeling, the GLP module strengthens shallow feature representations before fusion, and the SCE Block helps refine road boundaries and reduce fragmented predictions. These results indicate that GLP-Mamba provides an effective solution for fine-grained road segmentation in complex remote sensing scenarios.

  • Open Access

    OccPepSeg-YOLO for Instance Segmentation of Occluded Peppers in Field Images

    by Xinran Yu, Mingxi Jiang, Fei Gao, Yize Fan, Yanyan Bai, Zhigang Peng, Qi Lu, Qian Liu, Shengyong Xu

    Agricultural operations such as pepper harvesting, fruit counting, and field phenotyping rely on accurate visual recognition and instance segmentation algorithms. However, pepper fruits in complex field environments often exhibit slender and curved shapes, partial occlusion, ambiguous boundaries, and adhesion between adjacent instances. Existing object detection and instance segmentation methods therefore struggle to obtain complete fruit masks, which adversely affects subsequent fruit counting, contour measurement, and picking-point localization. To improve the instance segmentation accuracy of occluded peppers in complex field scenes, this study proposes OccPepSeg-YOLO, an improved model based on YOLO11n-seg. First, a P2FreqFusion module is introduced to fuse shallow, high-resolution detail features with deep semantic features, thereby enhancing the representation of fruit edges and tip regions. Second, an ASC module is designed to model the directional and scale-related morphological characteristics of pepper fruits, while a BoundaryGate module strengthens responses at occlusion interfaces and boundaries between adjacent instances. Finally, an OccPepSegment multi-scale prototype segmentation head is constructed, and a BDoU loss function is introduced to improve the boundary consistency of instance masks. Experiments on a self-constructed field-pepper instance segmentation dataset showed that OccPepSeg-YOLO achieved M-P, M-R, M-mAP50, and M-mAP50–95 values of 93.87%, 92.09%, 97.17%, and 82.31%, respectively, representing improvements of 5.59, 3.18, 3.83, and 9.52 percentage points over YOLO11n-seg. Further comparisons with representative YOLO-based instance segmentation models, including YOLOv8n-seg, YOLOv9c-seg, YOLO12n-seg, and YOLOv26n-seg, demonstrated that OccPepSeg-YOLO achieved the best overall segmentation performance. In particular, its M-mAP50–95 exceeded the best competing result obtained by YOLOv9c-seg by 8.35 percentage points. Under a unified repeated-inference protocol on an RTX 3090 GPU using FP32 precision, a batch size of 1, and 640 × 640 inputs, OccPepSeg-YOLO achieved a mean inference latency of 15.801 ± 1.238 ms, a P95 latency of 17.323 ms, and a throughput of 63.29 FPS. These results demonstrate that the proposed model can produce more complete pepper instance masks under leaf occlusion, fruit overlap, and complex background conditions, providing technical support for field-pepper recognition, fruit counting, and visual perception by agricultural robots.

  • Open Access

    A Two-Stage Deep Learning Method for Non-Invasive Sow Body Temperature Prediction Fusing Thermal Imaging and Environmental Parameters

    by Shengyong Xu, Ziyi Qin, Qiao Huang, Chen Tan, Xuewen Xu, Xuan Li

    Traditional rectal temperature measurement in pigs induces stress in animals, imposes a heavy labor burden on staff, and increases the risk of cross-infection. This study proposes a non-invasive deep learning approach to predict porcine rectal temperature by combining infrared thermal images of thermal windows with environmental parameters. A multimodal dataset is constructed by synchronously collecting thermal images, environmental parameters, and actual rectal temperatures. Mask Region-based Convolutional Neural Network (Mask R-CNN), You Only Look Once version 8 small (YOLOv8s), and YOLOv11s are employed to automatically detect or segment thermal window regions, from which the maximum temperature of each region is extracted. To enhance model generalization under varying environmental conditions, a two-stage hybrid regression framework is established. In this framework, a Convolutional Neural Network (CNN) extracts spatial features from thermal images, a fully connected network (FCNN) encodes regional surface temperatures and environmental parameters, and a Transformer module captures cross-modal dependencies to generate a preliminary prediction. Subsequently, a Random Forest (RF) regressor is applied for residual correction and final output optimization. Comparative experiments on single-region, dual-region, and triple-region combinations demonstrate that the “eye + vulva” dual-region scheme yields the optimal performance, with a mean absolute error (MAE) of 0.1796 °C and a coefficient of determination (R2) of 0.8212. The prediction error of this scheme is reduced by 42.3% compared with the best-performing unimodal model. The proposed method provides a fast, accurate, and stress-free solution for porcine body temperature monitoring, thereby supporting the development of intelligent health management in livestock farming.

  • Open Access

    Research on a Portable Multispectral Imaging System for Starch Content Detection in Watermelon–Pumpkin Grafted Seedling Leaves

    by Shengyong Xu, Honglei Yang, Yu Zeng, Shaodong Wang, Shuo Yang, Zhilong Bie, Yuan Huang

    Plant leaf starch content is a critical indicator of metabolic status, yet traditional enzymatic methods are destructive, labor-intensive, and costly. This study proposes a novel non-destructive detection method using watermelon–pumpkin grafted seedlings. To optimize hardware design, 12 characteristic wavelengths were identified via competitive adaptive reweighted sampling (CARS). A portable multispectral imaging system was developed, featuring narrowband LEDs and integrated human–computer interaction software for real-time visualization. We constructed a multimodal deep learning architecture that integrates a convolutional neural network (CNN) for spatial feature extraction from RGB images, a fully connected neural network (FCNN) for spectral data, and a Transformer network for high-level feature fusion. Experimental results showed that the ShuffleNet v2-Transformer model achieved an R2 of 0.956 (RMSE = 0.036) for watermelon leaves, while the EfficientNet b1-Transformer model reached an R2 of 0.967 (RMSE = 0.052) for pumpkin leaves. This multimodal approach significantly outperformed conventional PLSR and single-modal CNN models, demonstrating superior ability in processing long-range dependencies within spectral–spatial data. The system enables accurate detection with a throughput of 120 samples per hour at a hardware cost approximately 90% lower than commercial multispectral cameras. This provides an efficient, low-cost solution for large-scale monitoring of plant physiological indicators in precision breeding.

  • Open Access

    3D Point Cloud Segmentation Algorithm Based on Deep Learning and Its Application in Phenotype Detection

    by Bin Zhong, Liu Yang, Jing Lu, Shengyong Xu

    Three‐dimensional measurement technology based on point clouds can effectively solve the problem of plant occlusion and is a hot research direction for plant phenotyping methods. Rapid and low‐cost 3D reconstruction and accurate 3D point cloud segmentation are two major challenges in 3D phenotyping technology. Taking watermelon seedlings as an example, we proposed a new phenotyping method that uses the Instant‐NGP for 3D reconstruction and the improved PointNet++ for 3D point cloud segmentation. We performed dynamic downsampling and filtering based on the point cloud scale and designed different phenotypic measurement methods for hypocotyl and leaf point clouds. To overcome the difficulty of measuring the hypocotyl caused by slenderness, curvature and inclination, we proposed a segmented stem 3D point cloud skeleton extraction algorithm. The experimental results show that our method achieved satisfactory measurement results for the seedling phenotypes of four growth stages. The detection accuracy of the number of cotyledon leaves and the number of true leaves both exceed 95% and the coefficient of determination ( R 2 ) of leaf area, hypocotyl length and stem diameter phenotypes are all beyond 0.8. The proposed method provides a novel, efficient and precise 3D plant phenotyping solution, with good application and promotion value.

  • Open Access

    GCCA: Utilizing Gradient Matching for Open-set Domain Adaptation of Vision-language Models

    by Mingxi Jiang, Shengyong Xu

    Unsupervised Domain Adaptation (UDA) traditionally focuses on learning domain-invariant representations by aligning source and target domains, yet struggles in open-set scenarios where target domains contain unseen categories. Recent advances in vision-language models like CLIP, which leverage multimodal pre-training on large-scale image-text pairs, exhibit remarkable zero-shot generalization but still face performance degradation under significant domain shifts. In this work, we propose a gradient matching strategy to enhance CLIP's adaptability in open-set UDA. Unlike conventional prompt-based adaptation methods that independently optimize domain-specific parameters, our approach formulates domain alignment as a gradient consensus problem. Specifically, we enforce gradient direction consistency between source and target objectives to reconcile their optimization paths, while applying gradient norm penalization to prevent overfitting to domain-specific noise. This dual mechanism allows CLIP to preserve its inherent cross-modal discriminability while acquiring domain-robust features. Experiments across diverse open-set adaptation benchmarks demonstrate that our gradient matching framework substantially outperforms existing CLIP adaptation baselines, achieving state-of-the-art performance without compromising model stability.

  • Open Access

    A 3D reconstruction platform for complex plants using OB-NeRF

    by Sixiao Wu, Changhao Hu, Boyuan Tian, Yuan Huang, Shuo Yang, Shanjun Li, Shengyong Xu

    Introduction: Applying 3D reconstruction techniques to individual plants has enhanced high-throughput phenotyping and provided accurate data support for developing "digital twins" in the agricultural domain. High costs, slow processing times, intricate workflows, and limited automation often constrain the application of existing 3D reconstruction platforms.Methods: We develop a 3D reconstruction platform for complex plants to overcome these issues. Initially, a video acquisition system is built based on "camera to plant" mode. Then, we extract the keyframes in the videos. After that, Zhang Zhengyou's calibration method and Structure from Motion(SfM)are utilized to estimate the camera parameters. Next, Camera poses estimated from SfM were automatically calibrated using camera imaging trajectories as prior knowledge. Finally, Object-Based NeRF we proposed is utilized for the fine-scale reconstruction of plants. The OB-NeRF algorithm introduced a new ray sampling strategy that improved the efficiency and quality of target plant reconstruction without segmenting the background of images. Furthermore, the precision of the reconstruction was enhanced by optimizing camera poses. An exposure adjustment phase was integrated to improve the algorithm's robustness in uneven lighting conditions. The training process was significantly accelerated through the use of shallow MLP and multi-resolution hash encoding. Lastly, the camera imaging trajectories contributed to the automatic localization of target plants within the scene, enabling the automated extraction of Mesh. Results and discussion: Our pipeline reconstructed high-quality neural radiance fields of the target plant from captured videos in just 250 seconds, enabling the synthesis of novel viewpoint images and the extraction of Mesh. OB-NeRF surpasses NeRF in PSNR evaluation and reduces the reconstruction time from over 10 hours to just 30 Seconds. Compared to Instant-NGP, NeRFacto, and NeuS, OB-NeRF achieves higher reconstruction quality in a shorter reconstruction time. Moreover, Our reconstructed 3D model demonstrated superior texture and geometric fidelity compared to those generated by COLMAP and Kinect-based reconstruction methods. The $R^2$ was 0.9933,0.9881 and 0.9883 for plant height, leaf length, and leaf width, respectively. The MAE was 2.0947, 0.1898, and 0.1199 cm. The 3D reconstruction platform introduced in this study provides a robust foundation for high-throughput phenotyping and the creation of agricultural “digital twins”.


  • Open Access

    Design of a Non-Destructive Seed Counting Instrument for Rapeseed Pods Based on Transmission Imaging

    by Shengyong Xu, Rongsheng Xu, Pan Ma, Zhenhao Huang, Shaodong Wang, Zhe Yang, Qingxi Liao

    Pod counting of rapeseed is a critical step in breeding, cultivation, and agricultural machinery research. Currently, this process relies entirely on manual labor, which is both labor-intensive and inefficient. This study aims to develop a semi-automatic counting instrument based on transmission image processing and proposes a new algorithm for processing transmission images of pods to achieve non-destructive, accurate, and rapid determination of the seed count per pod. Initially, the U-NET network was used to segment and remove the stem and beak from the pod image; subsequently, adaptive contrast enhancement was applied to adjust the contrast of the G-channel image of the pod to an appropriate range, effectively eliminating the influence of different varieties and maturity levels on the translucency of the pod skin. After enhancing the contrast, the Sauvola algorithm was employed for threshold segmentation to remove the pod skin, followed by thinning and dilation of the binary image to extract and remove the central ridge lines, detecting the number and area of connected domains. Finally, the seed count was determined based on the ratio of each connected domain’s area to the mean area of all connected domains. A transmission imaging device that mimics the human eye’s method of counting seeds was designed, incorporating an LED transmission light source, photoelectric switch-triggered imaging slot, an industrial camera, and an integrated packaging frame. Human–machine interaction software based on PyQt5 was developed, integrating functions such as communication between upper and lower machines, image acquisition, storage, and processing. Operators simply need to place the pod in an upright position into the imaging device, where its transmission image will be automatically captured and processed. The results are displayed on a touchscreen and stored in Excel spreadsheets. The experimental results show that the instrument is accurate, user-friendly, and significantly reduces labor intensity. For various varieties of rapeseed pods, the seed counting accuracy reached 97.2% with a throughput of 372 pods/h, both of which are significantly better than manual counting and have considerable potential for practical applications.


  • Open Access

    Point clouds segmentation of rapeseed siliques based on sparse-dense point clouds mapping

    by Yuhui Qiao, Qingxi Liao, Moran Zhang, Binbin Han, Chengli Peng, Zhenhao Huang, Shaodong Wang, Guangsheng Zhou, Shengyong Xu

    In this study, we propose a high-throughput and low-cost automatic detection method based on deep learning to replace the inefficient manual counting of rapeseed siliques. First, a video is captured with a smartphone around the rapeseed plants in the silique stage. Feature point detection and matching based on SIFT operators are applied to the extracted video frames, and sparse point clouds are recovered using epipolar geometry and triangulation principles. The depth map is obtained by calculating the disparity of the matched images, and the dense point cloud is fused. The plant model of the whole rapeseed plant in the silique stage is reconstructed based on the structure-from-motion (SfM) algorithm, and the background is removed by using the passthrough filter. The downsampled 3D point cloud data is processed by the DGCNN network, and the point cloud is divided into two categories: sparse rapeseed canopy siliques and rapeseed stems. The sparse canopy siliques are then segmented from the original whole rapeseed siliques point cloud using the sparse-dense point cloud mapping method, which can effectively save running time and improve efficiency. Finally, Euclidean clustering segmentation is performed on the rapeseed canopy siliques, and the RANSAC algorithm is used to perform line segmentation on the connected siliques after clustering, obtaining the three-dimensional spatial position of each silique and counting the number of siliques. The proposed method was applied to identify 1457 siliques from 12 rapeseed plants, and the experimental results showed a recognition accuracy greater than 97.80%. The proposed method achieved good results in rapeseed silique recognition and provided a useful example for the application of deep learning networks in dense 3D point cloud segmentation.


  • Open Access

    Early Identification and Localization Algorithm for Weak Seedlings Based on Phenotype Detection and Machine Learning

    by Shengyong Xu, Yi Zhang, Wanjing Dong, Zhilong Bie, Chengli Peng, Yuan Huang

    It is important to propose the correct decision for culling and replenishing seedlings in factory seedling nurseries to improve the quality of seedlings and save resources. To solve the problems of inefficiency and subjectivity of the existing traditional manual culling and replenishment of seeds, this paper proposes an automatic method to discriminate the early growth condition of seedlings. Taking watermelon plug seedlings as an example, Azure Kinect was used to collect data of its top view three times a day, at 9:00, 14:00, and 19:00. The data were collected from the time of germination to the time of main leaf growth, and the seedlings were manually determined to be strong or weak on the last day of collection. Pre-processing, image segmentation, and point cloud processing methods were performed on the collected data to obtain the plant height and leaf area of each seedling. The plant height and leaf area on the sixth day were predicted using an LSTM recurrent neural network for the first three days. The R squared for plant height and leaf area prediction were 0.932 and 0.901, respectively. The dichotomous classification of normal and abnormal seedlings was performed using six machine learning classification methods, such as random forest, SVM, and XGBoost, for day six data. The experimental results proved that random forest had the highest classification accuracy of 84%. Finally, the appropriate culling and replenishment decisions are given based on the classification results. This method can provide some technical support and a theoretical basis for factory seedling nurseries and transplanting robots.