GAN-Based Deblurring Algorithm for Drug Package Detection

With the development of Industry 4.0, many manufacturers are moving in the direction of automation and intelligence. Drug package inspection quality testing methods have also changed from traditional manual inspection to machine vision inspection. However, due to the long exposure time or vibration of the camera, the images are blurred, which will seriously affect the accuracy of the defect detection. We propose a deblur model with higher level of image restoration and more lightweight. Using reserving and merging blocks to replace the original reserving and merging blocks, which can be used in the inference phase to turn ResBlocks into equivalent straight structures using reserving and merging operations, making the network lighter. The pyramid squeeze attention module is also incorporated to enable the model to efficiently extract finer-grained multi-specimen scale spatial information, thus making more efficient use of feature space information. Finally, a drug package dataset was produced and the model was trained to obtain good deblurring results. The comparative experiments confirmed that the model has better image reduction and excellent reduction speed, and can be further cited in production inspection pipelines to improve detection accuracy.