A high-efficiency dirty-egg detection system based on YOLOv4 and TensorRT

Abstract

Eggshells are easily stained with dirt and the dirty eggs are more likely to be contaminated with bacteria, which has adverse effects on the health of consumers. The traditional methods to filter out the dirty eggs are mainly based on image processing, which suffers low accuracy that results in the need for manual screening in many cases. The paper proposes an embedded system based on artificial intelligence to detect dirty spots on eggshells. The proposal adopts the neural network YOLOv4 to detect dirty eggs and accelerates the detection process by utilizing the TensorRT, and finally, the system is implemented on Jetson Nano. The system achieves a high speed with no accuracy loss. Experimental results show that the system can identify multiple dirty spots of multiple eggs within one image, the accuracy reaches 75.88% and the speed is up to 2.3 fps. The system offers a promising solution in the field of dirty spots detection for eggs.