Yibing Zhang
Yibing Zhang has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Dongfang Zhang
Ministry of Education of China-Hebei Province Joint Innovation Center for Efficient Green Vegetable, Key Laboratory of Vegetable Germplasm Innovation and Utilization of Hebei, College of Horticulture, Hebei Agricultural University, Baoding 071000, China
2 shared publications - Jun Zhang
School of mechanical and electrical engineering, Hebei Agricultural University, Baoding 071000, China
2 shared publications - Xiaofei Fan
School of mechanical and electrical engineering, Hebei Agricultural University, Baoding 071000, China
2 shared publications - Yuhong Zhou
School of mechanical and electrical engineering, Hebei Agricultural University, Baoding 071000, China
2 shared publications - Jiaqi Wang
School of mechanical and electrical engineering, Hebei Agricultural University, Baoding 071000, China
1 shared publication - Jiaqi Wang
School of mechanical and electrical engineering, Hebei Agricultural University, Baoding 071000, China
1 shared publication
Latest Publications
- Open Access
Early Detection of Eggplant Verticillium Wilt Under Magnetized Water Irrigation Using Multi-Source Data Fusion and Ensemble Learning
Verticillium wilt is a major soil-borne disease affecting eggplant production. Early detection before symptom manifestation can reduce yield losses. Compared with traditional destructive physiological assays, sensor-based technologies enable efficient, nondestructive monitoring. However, magnetized water irrigation may reduce phenotypic separability between healthy and diseased plants, complicating disease detection using single-source data. This study developed a multisource data fusion strategy that integrates multispectral reflectance, chlorophyll fluorescence parameters, and Soil Plant Analysis Development (SPAD) values to construct a weighted ensemble model combining extreme gradient boosting (XGBoost) and support vector machine (SVM). Based on a 28-dimensional multisource feature set collected one to seven days post-inoculation (DPI), the model achieved an overall accuracy of 88.1% and an area under the receiver operating characteristic curve (AUC) of 0.925, with accuracies of 90.1% and 86.0% under ordinary and magnetized water irrigation, respectively. Ablation experiments showed that spectral features alone achieved a recall of only 62.9% under magnetized water irrigation, whereas multisource fusion increased recall to 73.2%. Compared with visual inspection, which requires approximately 10 days for disease confirmation, the model identified diseased plants as early as DPI 3, achieving 91.8% accuracy and an AUC of 0.978. Overall, multisource data fusion improved early Verticillium wilt detection across irrigation regimes. This study demonstrates the novelty of utilizing this fusion strategy to overcome the reduced phenotypic separability induced by magnetized water irrigation, providing a robust and nondestructive diagnostic tool for agricultural disease management.
Open AccessEarly Detection of Eggplant Verticillium Wilt Under Magnetized Water Irrigation Using Multi-Source Data Fusion and Ensemble Learning
Verticillium wilt is a major soil-borne disease affecting eggplant production. Early detection before symptom manifestation can reduce yield losses. Compared with traditional destructive physiological assays, sensor-based technologies enable efficient, nondestructive monitoring. However, magnetized water irrigation may reduce phenotypic separability between healthy and diseased plants, complicating disease detection using single-source data. This study developed a multisource data fusion strategy that integrates multispectral reflectance, chlorophyll fluorescence parameters, and Soil Plant Analysis Development (SPAD) values to construct a weighted ensemble model combining extreme gradient boosting (XGBoost) and support vector machine (SVM). Based on a 28-dimensional multisource feature set collected one to seven days post-inoculation (DPI), the model achieved an overall accuracy of 88.1% and an area under the receiver operating characteristic curve (AUC) of 0.925, with accuracies of 90.1% and 86.0% under ordinary and magnetized water irrigation, respectively. Ablation experiments showed that spectral features alone achieved a recall of only 62.9% under magnetized water irrigation, whereas multisource fusion increased recall to 73.2%. Compared with visual inspection, which requires approximately 10 days for disease confirmation, the model identified diseased plants as early as DPI 3, achieving 91.8% accuracy and an AUC of 0.978. Overall, multisource data fusion improved early Verticillium wilt detection across irrigation regimes. This study demonstrates the novelty of utilizing this fusion strategy to overcome the reduced phenotypic separability induced by magnetized water irrigation, providing a robust and nondestructive diagnostic tool for agricultural disease management.
- Open Access
DSS-YOLO: A Lightweight Flower and Stamen Detection Model for Greenhouse Tomato Pollination Assistance
The accurate detection of tomato flowers and stamens in greenhouse environments is a key technology for achieving automated pollination and intelligent crop management. However, during the detection process, tomato flowers and stamens often face challenges such as overlapping, leaf occlusion, complex backgrounds, and difficulties in small-scale feature extraction, which severely impact detection accuracy. Therefore, this paper proposes a lightweight detection model, termed DSS-YOLO. First, the backbone network of YOLOv11n was replaced with HGNetv2, and depthwise separable convolution (DWConv) was incorporated to construct a multiscale lightweight feature extraction network, named DWHGNetv2. This design enhances the feature extraction capability for tomato flowers while reducing computational cost and overall model complexity. Second, traditional convolution-based downsampling layers were replaced with the SCDown module to improve computational efficiency and feature representation. Furthermore, the SIoU loss function was introduced to optimize the localization accuracy of angle-sensitive targets, such as stamens. Experimental results demonstrate that DSS-YOLO consistently outperforms the baseline YOLOv11n. Compared with YOLOv11n, the model size, parameter count, and computational cost are reduced by 34%, 36%, and 35%, respectively, while precision, recall, and mean Average Precision at an IoU threshold of 0.5 (mAP@0.5) are improved by 1.1%, 1.0%, and 0.7%, respectively. These results indicate that DSS-YOLO provides a robust and efficient solution for high-precision tomato flower and stamen detection, meeting the practical requirements of pollination robots in greenhouse environments.
Open AccessDSS-YOLO: A Lightweight Flower and Stamen Detection Model for Greenhouse Tomato Pollination Assistance
The accurate detection of tomato flowers and stamens in greenhouse environments is a key technology for achieving automated pollination and intelligent crop management. However, during the detection process, tomato flowers and stamens often face challenges such as overlapping, leaf occlusion, complex backgrounds, and difficulties in small-scale feature extraction, which severely impact detection accuracy. Therefore, this paper proposes a lightweight detection model, termed DSS-YOLO. First, the backbone network of YOLOv11n was replaced with HGNetv2, and depthwise separable convolution (DWConv) was incorporated to construct a multiscale lightweight feature extraction network, named DWHGNetv2. This design enhances the feature extraction capability for tomato flowers while reducing computational cost and overall model complexity. Second, traditional convolution-based downsampling layers were replaced with the SCDown module to improve computational efficiency and feature representation. Furthermore, the SIoU loss function was introduced to optimize the localization accuracy of angle-sensitive targets, such as stamens. Experimental results demonstrate that DSS-YOLO consistently outperforms the baseline YOLOv11n. Compared with YOLOv11n, the model size, parameter count, and computational cost are reduced by 34%, 36%, and 35%, respectively, while precision, recall, and mean Average Precision at an IoU threshold of 0.5 (mAP@0.5) are improved by 1.1%, 1.0%, and 0.7%, respectively. These results indicate that DSS-YOLO provides a robust and efficient solution for high-precision tomato flower and stamen detection, meeting the practical requirements of pollination robots in greenhouse environments.