The article describes a sensor developed on the basis of a transformer-type transformer intended for studying metal films. The features of the combined use of the eddy current method and computer vision and neural network technologies for automatic detection of defects in thin metal films are considered. To automate the determination of the defective area of the objects under study, an ultra-precise neural network was implemented and trained on the U-net architecture, which was trained on 50 real research results and 50 sets obtained by random rotations. The design of an automatic eddy current system is considered, including a control unit, a generator, a positioning system and an eddy current sensor. The results of using the developed system and the trained neural network for detecting defects in thin copper films are presented. The experiments can be successfully used to detect discontinuities in thin metal films.