Journal ID : AMA-16-11-2021-10844
[This article belongs to Volume - 52, Issue - 03]
Total View : 495

Title : RAPID RECOGNITION OF SPROUTING POTATO IMAGES BASED ON IMPROVED YOLO V5

Abstract :

Sprouted potato detection is an essential measure before potatoes enter warehouse storage and can effectively reduce the chance of warehouse spoilage of potatoes. How to detect potato health intelligently and efficiently is important to improve the quality of potatoes before they enter the warehouse. To achieve the detection and grading of sprouted potatoes in a variety of complex scenarios, this study proposes a sprouted potato detection algorithm based on an improved YOLOV5 model. The Cross Conv module with improved feature similarity is used to replace the Conv of the original C3 module of YOLOV5, which improves the similarity loss problem in the fusion process and increases the feature expression capability; the SPPF with accelerated space pyramid is used instead of SPP for fusion pooling, which reduces the number of fusion parameters and accelerates the speed of fusion pooling; the 9-Mosaic algorithm is enhanced and optimized to strengthen small target features before the image enters Backbone; then the accuracy is further improved using hyperparametric evolution with genetic evolution anchor points and multi-scale training mechanism, and the experimental results show that: the improved model recognition accuracy reaches a minimum of 90.14%, the average accuracy of the whole class mAP@.5 reaches 88%, and the F1-score is 84%, which is higher than the original YOLOV5 network in the same test dataset The model mAP@.5 index is 7.4% higher than the original model under the same test dataset, which has obvious advantages over the existing model. The real-time sprouting potato image recognition based on improved YOLOV5 proposed in this study has good accuracy and effectiveness, which can basically meet the requirements for establishing automatic potato sorting line and realizing high-throughput and fast potato classification, and provide technical reference for intelligent agricultural equipment in modern agricultural environment

Full article