LBSR-YOLO: Blueberry health monitoring algorithm for WSN scenario application

• Published in: Computers and electronics in agriculture Vol. 238; p. 110803
• Main Authors: Song, Zhiwen, Li, Wei, Tan, Wei, Qin, Tao, Chen, Changsheng, Yang, Jing
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2025
• Subjects: Blueberry health monitoring; BSRN; Mobile deployment; Wireless sensor network; YOLO v10n; Wireless sensor network; Blueberry health monitoring; Mobile deployment; YOLO v10n; BSRN
• ISSN: 0168-1699
• DOI: 10.1016/j.compag.2025.110803

•The LBSR-YOLO is proposed for intensive blueberry health monitoring in WSN scenarios.•The proposed CSFPC module effectively reduces the complexity of the algorithm.•The BSRN algorithm improves the detection accuracy and prolongs the survival time of WSN nodes•Ablation experiments show that CCFF and DWR modules are helpful for the detection of small target blueberry diseases. Aiming at the problem of high cost and low efficiency of blueberry health monitoring in the blueberry intensive planting scene, present an enhanced LBSR-YOLO algorithm, combining the BSRN algorithm and YOLO v10n algorithm, for health monitoring of blueberries in a wireless sensor network (WSN) environment. Firstly, the input layer of the BSRN network uses large kernel weight sharing convolution (LKWSConv) to improve the quality of image generation, and partial convolution (PConv) is integrated into the network backbone to reduce complexity and parameters. Secondly, on the YOLOv10n network, free partial convolution (FPConv) is introduced to reduce model complexity and parameter count, and omni-dimensional dynamic convolution (ODConv) is introduced to enhance the model’s feature extraction capability. Cross-scale feature fusion module (CCFF) and dilation-wise residual module (DWR) are used to optimize and reconstruct the neck structure to detect small target objects. Finally, the improved network embedding is integrated into the LBSR-YOLO algorithm, and it is run on the edge computing node in the WSN application scenario. The results show that LBSR-YOLO achieves 79.3 % mAP on low-resolution images, with a model size of 3.7 MB and a computational complexity of 44.5 GFLOPs. For the edge detection node of the WSN scene, the detection of the same resolution image, the LBSR-YOLO algorithm can save 0.76 J of energy compared to the traditional method. Under the power supply of a 7.4 Wh battery, the LBSR-YOLO algorithm extends the battery life by 1300 s compared to the traditional method. The LBSR-YOLO algorithm can be deployed on low-cost devices to monitor blueberries’ health efficiently and has high use value.