Development of artificial vision with neural networks for identification of imperfections in tuna packaging with three cameras

Abstract

The objective of this research is to describe the development of artificial vision with neural networks to identify imperfections in tuna cans using three cameras. Python and specialized libraries such as TensorFlow, OpenCV, NumPy and Keras were used to develop the system. A convolutional neural network (CNN) was implemented with labeled images of cans in good and bad condition. The system uses three cameras to inspect cans from multiple angles and applies filters such as Gaussian blur, adaptive thresholding, and Canny algorithm to improve image quality and edge detection accuracy. An accuracy of 98.17% was achieved in training and 98.76% in image validation. It is recommended to perform additional testing with larger data sets, optimize the edge detection algorithm, explore other neural network architectures, and consider integration with robotics to automate the inspection process. Finally, it is concluded that the developed system is an effective tool to detect dents in tuna cans, improving quality control, reducing costs and guaranteeing the safety of the final product in the food industry.