In the food industry, products are often packed on trays or pallets in 2D structures on automated lines where damages can occur. The problem of detecting bad products was solved using computer vision combined with deep learning approach. Convolutional neural network was trained unsupervised due to the high variability of possible errors and the nature of industrial production. It was found that the system for automatic visual control of the products works well and successfully separates good and bad samples. We tested the system on stationary and moving objects traveling along the conveyor belt.
|
