Application of object detection algorithm based on deep learning in classification of wild ginseng grades

Abstract

Abstract Object detection based on deep learning has an excellent effect on wild ginseng classification. In recent years, the Object detection algorithm has been widely used in the classification research of various leaf plants, while the field of wild ginseng generally uses relatively backwards manual identification methods. The identification of wild ginseng by artificial methods are of low efficiency, subjectivity and poor accuracy. In order to solve this problem, this paper proposes a new structure based on PP-YOLO tiny object detector, namely PP-YOLO tiny BiFPN*, improving the backbone network and replacing its neck with the bidirectional feature pyramid network to enhance feature extraction. Combined with embedded devices, Paddle-Lite framework and OpenCV technology, an auxiliary wild ginseng grade classification system based on the deep learning algorithm and embedded devices is designed and implemented. In the test environment, the PP-YOLO tiny BiFPN* model obtained in the first and second grade wild ginseng classification precision of 99.61% and 99.72%, recall of 99.72% and 99.61%, the classification accuracy of the whole wild ginseng sample has reached 99.66%. The model size is only 2.27MB, and the processing speed of each image is 0.266 seconds per image, which is suitable for working in the environment of limited resources, and opens up a new idea for the intelligence classification of wild ginseng.