农机化研究

为解决苹果采摘机器人检测算法存在的网络结构复杂和参数量大的问题，提出一种基于YOLOv5的轻量化苹果检测算法。首先，将YOLOv5主干网络替换为MobileNetv3,为降低网络的计算复杂度，将深度可分离卷积引入到特征融合网络中；然后，在网络的关键位置引入注意力机制，以提高算法对苹果不同特征的提取能力；最后，使用CIoU作为改进网络的损失函数，以提升模型的检测效果。试验结果表明：改进模型的检测精度为91.5%,相较于SSD、Faster R-CNN,检测精度分别提高了2.35%、3.07%,相比于YOLOv5s检测精度提高了8.20%,且模型大小约为YOLOv5s的1/3。

A lightweight apple detection algorithm based on YOLOv5 was proposed to solve the problems of complex network structure and large parameter quantity in the detection algorithm of apple picking robots. Firstly, the YOLOv5 backbone network was replaced with MobileNetv3. To reduce the computational complexity of the network, deep separable convolution was introduced into the feature fusion network.Then, attention mechanisms were introduced at key locations in the network to improve the algorithm′s ability to extract different features of apples. Finally, CIoU was used as the Loss function of the improved network to improve the detection effect of the model. The test results showed that the detection accuracy of the improved model was 91.5%, which was 2.35% and 3.07% higher than SSD and Faster R-CNN, respectively.Compared to YOLOv5s, the detection accuracy had been improved by 8.20%, and the model size was about one-third of YOLOv5s.