基于Des-YOLO v4的复杂环境下苹果检测方法

Apple Detection Method in Complex Environments Using Des-YOLO v4

为使采摘机器人快速准确检测出复杂环境中的苹果,提出一种Des-YOLO v4算法与苹果检测方法。由于YOLO v4的网络结构复杂,提出一种Des-YOLO结构,可减少网络参数并提高算法的检测速度;在训练阶段,正负样本的不平衡会导致苹果误检,提出一种基于AP-Loss的类别损失函数,以提高苹果识别的准确性。通过自制的苹果数据集测试了Des-YOLO v4算法,并在苹果采摘机器人样机上完成了采摘实验。实验结果表明:Des-YOLO v4算法对苹果图像的平均精度值为93.1%,检测速度为53f/s;机器人单次采摘时间为8.7s,采摘成功率达92.9%,具有检测精度高、速度快等优点。

In order to make the harvesting robot detect apples quickly and accurately in complex environment,a Des-YOLO v4 algorithm and a detection method of apples are proposed.The network structure of YOLO v4 is complex,so a Des-YOLO structure based on the YOLO v4 network structure is proposed to improve the speed and accuracy of apple detection.To solve the imbalance of positive and negative samples in the collected data,a class loss function based on APLoss is proposed to improve the accuracy of apple recognition.The proposed algorithm is tested on the self-made apple image data set.The results show that the Des-YOLO v4 has ideal features with a mAP of apple detection of 97.13% and a detection speed of 53 f/s.Finally,the self-made apple-harvesting robot is used to carry out the picking experiment in the laboratory environment.The experimental results show that the picking time is 8.7 seconds and the successful picking rate of the robot is 92.9%.Therefore,the apple detection method proposed has the advantages of higher recognition accuracy and faster recognition speed.