Design and Implementation of Image Recognition for Package Crack of Dry-type Air-core Reactor Package

To deal with image recognition for package of dry type reactor crack, An improved method based on yolov3 neural network and visual saliency is proposed to identify cracks in complex environment. First of all, the defect location of package crack is carried out by using yolov3 algorithm; This paper presents an improved visual saliency method for crack detection with complex background, images are preprocessed by SLIC algorithm and Gaussian noise reduction algorithm, finally, the defect area is located after the saliency algorithm is used to calculate the processed image. experimental results show that the detection results of this method are suitable for actual scene.