Scratch Detection of the PET Bottle Preform Based on Deep Learning

Scratches are a common phenomenon in the production of the PET bottle preform, and traditional inspection by human eyes bring troubles to the automatic production process. In this paper, deep learning algorithm was used to detect PET bottle preform scratches. The detection algorithm based on YOLOv4 neural network was proposed. The detection effect was tested under different scratch degrees, which provide accurate information for automatic production picking of the PET bottle preform. The average detection time of a single image was 0.154s, and the shortest detection time was 0.133s.