Abstract: Tomato fruit detection is a key issue that needs to be addressed in order to achieve mechanization and automation of tomato harvesting. In response to the complex background, dense fruit distribution, and leaf obstruction in the tomato growth environment, an optimized YOLOv7 mature tomato recognition model is proposed. Based on the YOLOv7 model, the main network's ELAN module is replaced with a P-ELAN module to reduce network parameters and computational load, while enhancing the network's feature extraction capability. Additionally, an LSK attention mechanism is added in front of the detection head to dynamically adjust the receptive field using the feature extraction module, more effectively handling the differences in background information required for different targets. Finally, the EIoU loss function is introduced to more effectively guide the model in learning more accurate bounding box predictions, thereby accelerating the convergence of prediction boxes and improving the regression accuracy of prediction boxes. The improved algorithm not only has high recognition accuracy but is also more lightweight, making it well-suited for application in mature tomato harvesting scenarios.
崔世磊, 孙明革, 高聪, 郭晓龙, 李迎岗. 基于YOLOv7的番茄检测算法优化与实现[J]. 吉林化工学院学报, 2024, 41(3): 25-30.
CUI Shilei, SUN Mingge, GAO Cong, GUO Xiaolong, LI Yinggang. Optimization and Implementation of Tomato Detection Algorithm based on YOLOv7 . Journal of Jilin Institute of Chemical Technology, 2024, 41(3): 25-30.
