Defect Detection of Rubber Gloves Based on Normal Samples

Rubber gloves are widely favored in various industries, and the defect detection link in production is very important. This paper proposes a method for detecting defects in rubber gloves based on normal samples (no defects). First, random noise is added to the collected rubber glove images, and the size is standardized and unified. Secondly, establish an improved GANomaly network model. According to the small and difficult defect to distinguish characteristics of black spots and oil defects of rubber gloves, a noise-reducing convolutional autoencoder is built in the network model as a generation network, and the least square loss is introduced for the model of confrontation training. The model only uses normal sample input for training, learning the characteristic distribution of the normal sample and performs image reconstruction, and judges whether it has defects according to the reconstruction effect score of the input sample during testing. The experiment is based on the rubber glove image data set, and the results show that the method can accurately identify whether there are defects on the rubber glove.