When the bubbles of medical plastic bottles are detected, the arbitrariness of the bubble position in the bottle body, the uncertainty of the bubble size, and the similarity between the bubble characteristics and the bottle body characteristics increase the difficulty of detecting the bubble defects. In order to solve the above problems in the detection of bubble defects, a real-time segmentation algorithm based on improved Fast Segmentation Convolutional Neural Network (Fast-SCNN) was proposed. The basic framework of the segmentation algorithm is the Fast-SCNN. In order to make up for the lack of robustness of the original network segmentation scale, the ideas of the usage of the information between the channels of Squeeze-and-Excitation Networks (SENet) and the multi-level skip connection were adopted. Specifically, the deep features were extracted by further down-sampling of the network, the up-sampling operation was merged with SELayer module in the decoding stage, and the skip connections with the shallow layer of the network were increased two times at the same time. Four sets of experiments were designed for comparison on the bubble dataset with the Mean Intersection over Union (MIoU) and the segmentation time for single image of the algorithm used as evaluation indicators. The experimental results show that the comprehensive performance of the improved Fast-SCNN is the best, this network has the MIoU of 97.08%, the average segmentation time for a medical plastic bottle of 24.4 ms, and the boundary segmentation accuracy 2.3% higher than Fast-SCNN, which improves the segmentation ability of tiny bubbles, and this network has the MIoU improved by 0.27% and the time reduced by 7.5 ms compared to U-Net, and the comprehensive detection performance far better than Fully Convolutional Networks (FCN-8s). The proposed algorithm can effectively segment smaller bubbles with unclear edges and meet the engineering requirements for real-time segmentation and detection of bubble defects.

Focusing on the problems such as low inspection speed and low accuracy caused by subjective factors in the manual inspection method of ampoule packaging quality, an inspection algorithm based on machine vision and lightweight neural network was proposed. First, threshold processing, tilt correction and cutting of ampoule regions were performed on the images to be inspected by using the threshold segmentation and affine transformation methods in machine learning. Second, the network structure of the classification algorithm was designed according to the characteristics of images and the requirements of defect recognition. Finally, the ampoule packaging defect dataset was constructed by collecting the images of the production site. After that, the proposed ampoule packaging defect identification network was verified, and the accuracy and inspection speed of the algorithm deployed on the Jetson Nano embedded platform were tested. Experimental results show that, taking the product of five ampoules each box as the example, the proposed ampoule packaging quality inspection algorithm takes 70.1 ms/box averagely, that is up to 14 boxes/s, and has the accuracy of 99.94%. It can achieve online high-precision ampoule packaging quality inspection on the Jetson Nano embedded platform.