基于YOLOv7的番茄检测算法优化与实现

Optimization and Implementation of Tomato Detection Algorithm based on YOLOv7

番茄果实目标检测是实现番茄采摘机械化、自动化需要解决的关键问题。针对番茄生长环境中背景复杂、果实密集、枝叶遮挡等问题,提出一种优化的YOLOv7成熟番茄识别模型。该模型在YOLOv7模型基础上,首先将主干网络中的ELAN模块用P-ELAN模块替换,降低了网络参数量与计算量,同时提升网络的特征提取能力。其次,在检测头前加入LSK注意力机制,利用特征提取模块动态调整感受野,更有效地处理了不同目标所需的背景信息差异。最后,引入EIoU损失函数,能够更有效地引导模型学习更准确的边界框预测,从而加速预测框的收敛、提高预测框的回归精度。改进后的算法不但识别精度高,同时更为轻量化,可以较好地应用于成熟番茄的采摘场景。

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 was proposed.Based on the YOLOv7 model,the main network's ELAN module was 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 was 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 was 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.