基于改进YOLOv5的温室番茄果实检测算法

Greenhouse Tomato Fruit Detection Algorithm Based on Improved YOLOv5

为实现番茄采摘机器人在温室环境下准确高效地作业,提出了一种基于改进YOLOv5深度学习网络的番茄果实检测算法——SL-YOLOv5。改进的算法中,通过SimSPPF(Simplified Spatial Pyramid Pooling-Fast)替换SPP(Spatial Pyramid Pooling)网络结构,加快网络对番茄果实的检测速度。同时,引入大型分离卷积注意力机制LSKA(Large Separable Kernel Attention),以此提高网络对番茄果实的检测精度;在自制的温室番茄果实数据集上对该算法进行了测试。结果表明:改进后的SL-YOLOv5算法对成熟番茄果实识别准确率达到0.963,对未成熟番茄果实识别准确率达到0.951,综合识别准确率达到0.957,高于原YOLOv5算法2.4%;改进后的SL-YOLOv5算法检测速度FPS(Frames Per Second)达到6.75,比原YOLOv5算法提高17.2%。改进后的SL-YOLOv5算法能够在温室环境下准确检测出番茄果实,并能够对成熟番茄和未成熟番茄进行分类,与原YOLOv5算法相比,有效提升了检测的精度和速度。

In order to realize the accurate and efficient operation of tomato picking robot in greenhouse environ-ment,a tomato fruit detection algorithm based on improved YOLOv5 deep learning network,SL-YOLOv5,was proposed.In the improved algorithm,the SPP(Spatial Pyramid Pooling)network structure is replaced by SimSPPF(Simplified Spatial Pyramid Pooling-Fast)to speed up the detection speed of tomato fruits,and the large separable kernel attention(LSKA)is introduced to improve the detection accuracy of tomato fruits.The algorithm was tested on a self-made greenhouse tomato fruit dataset,and the detection speed and accuracy of the network before and after the improvement were compared.The experimental results show that the improved SL-YOLOv5 algorithm has an accuracy rate of 0.963 for the recognition of ripe tomato fruits,0.951 for unripe tomato fruits,and 0.957 for the comprehensive recognition in the greenhouse environment,which is 2.4%higher than that of the original YOLOv5 algorithm,and the detection speed(FPS)is 6.75,which is 17.2%higher than that of the original YOLOv5 algorithm.The improved algorithm can detect tomato fruits in the greenhouse environment,classify ripe tomatoes and immature tomatoes,and effectively improve the speed and accuracy of tomato fruit detection in the greenhouse environment.