Apple object detection in natural environment based on YOLO v5

摘要

Apple's recognition and positioning is an important part of the picking robot system is also a key function. Apple target recognition in natural environment is affected by environmental factors such as weather, branches and leaves, which increase the difficulty of apple recognition and reduce the accuracy of recognition. In recent years, artificial intelligence computer vision technology has been able to accurately and quickly detect objects through continuous development. Based on deep learning, an apple detection method in natural environment based on YOLO v5s deep learning was proposed. Images of apple in natural environment were tagged and sent to the YOLO v5s target detection network with SE attention mechanism for network model training. After multiple rounds of network model training, the accuracy rate of the final model is 86.0%, the recall rate is 93.6%, the average accuracy is 90.9%, the model size is 15.6MB, and the detection speed is 84.0 frames /s. The results show that YOLO v5s-SE can accurately and quickly realize the target detection of apple in natural environment, and the model has high robustness and relative fast recognition speed.