Deep Learning Based Real-time Visual Inspection for Harvesting Apples

With the rapid development of artificial intelligence, deep learning has been widely used in agricultural production, such as harvesting apples. However, distinguishing damaged apples from good apples is still labor-intensive. To address this problem, we propose a deep learning based real-time visual inspection system for harvesting apples. The system is quick and accurate in distinguishing between good and damaged apples. In detail, we create an apple dataset for inspection. Meanwhile, the YOLOv4 model is applied for real-time detection. The experimental results show the effectiveness of our method with 93% of F1, 98.34% of precision, 88.12% of recall and 88.09% of AP on one of the objects. Therefore, the study in this paper provides a good application in agricultural production.