Potato Surface Defect Detection Based on Deep Transfer Learning

Food defect detection is crucial for the automation of food production and processing. Potato surface remains challenging due to irregular shape potato individuals various types defects. This paper employs deep convolutional neural network (DCNN) models detection. In particular, we applied transfer learning by fine-tuning a base model through three DCNN models—SSD Inception V2, RFCN ResNet101, Faster RCNN ResNet101—on self-developed dataset, achieved an accuracy 92.5%, 95.6%, 98.7%, respectively. ResNet101 presented best overall performance in speed accuracy. It was selected as final out-of-sample testing, further demonstrating model’s ability generalize.