Image Enhancement Algorithm based on Local Contrast for Convolutional Neural Network-Based Infrared Target Recognition: *Note: Subtitles are not captured in Xplore and should not be used

In the task of infrared weak and small target recognition, in order to improve the image quality and solve the problem of poor learning ability of convolutional neural network (CNN) due to the imbalance of positive and negative samples, this study proposes a three-stage image enhancement algorithm named adaptive filter for quality enhancement based on local contrast (AFQELC). Firstly, AFQELC analyzes the statistical characteristics of the image and constructs a local contrast adaptive filter (LCAF) to enhance the detailed information of the image, which promotes the deep learning model to learn low-level semantic information. Secondly, principal component analysis (PCA) fusion combines information from the original image and the image enhanced by LCAF to reduce noise. Finally, the gradient component of the original image is extracted to further revise the processing result, which promotes the deep learning model to learn advanced semantic information. The proposed non-data-driven algorithm has a clear and interpretable process, which is superior to other traditional and neural network based image enhancement algorithms. The experiment results show that the quality of images enhanced by AFQELC is improved significantly. In addition, AFQELC can improve the recognition and positioning accuracy of the CNN-based algorithm for infrared target recognition by alleviating the imbalance of positive and negative samples.