Application of Object Detection Algorithm Based on Deep Learning in Classification of Wild Ginseng Grades

Deep learning-based object detection has shown significant efficacy in the classification of wild ginseng. However, the field of wild ginseng classification predominantly relies on outdated manual identification methods, resulting in low efficiency, subjective judgments, and poor accuracy. To address this issue, this study proposes a novel architecture based on the PP-YOLO tiny object detector, specifically PP-YOLO tiny BiFPN*, which improves the backbone network and replaces its neck with BiFPN to enhance feature extraction. Leveraging embedded devices, the Paddle-Lite framework, and OpenCV technology, an auxiliary system for wild ginseng grade classification is designed and implemented using deep learning algorithms. In the experimental setting, the PP-YOLO tiny BiFPN model achieved precision rates of 99.61% and 99.72%, and recall rates of 99.72% and 99.61% in classifying first and second-grade wild ginseng, respectively. The overall classification accuracy for the entire wild ginseng dataset reached 99.66%. Remarkably, the model size is only 2.27MB, and it processes each image at a speed of 0.266 seconds per image. Consequently, it is highly suitable for resource-constrained environments, providing a novel approach to intelligent classification of wild ginseng.