An Image Distortion Correction Algorithm Based on Feature Fusion AlexNet with Wide First-layer Kernels

Intelligent power inspection based on image is the development direction to improve the efficiency of power measurement task, and image distortion caused by lens parameters during data collection can affect the efficiency of power inspection. This paper proposes an image distortion correction algorithm based on Feature Fusion AlexNet with Wide First-layer kernels (FFAWNet). This method enhances the model’s ability to capture image distortion features by fusing high-resolution fine-grained features in the low-level features and high-level semantic features. Firstly, the lens parameters are used to fit the functional relationship between distortion parameters, generating inspection images within the common radial distortion range. Then, FFAWNet is used to learn the features of radial distortion, and the input image is mapped as distortion parameters to achieve image distortion correction. The experiment proves that the correction error of the FFAWNet method proposed in this paper is within 1 pixel.