Low-Resolution Infrared Small Target Detection Algorithm Based on Superpixel Segmentation Technology and AdaBoost-SVM Algorithm

This paper presents a novel and effective method which fuses superpixel segmentation technology and AdaBoost Support Vector Machine (AdaBoost-SVM) algorithm for low-resolution infrared small target detection in very complex backgrounds. First, a simple linear iterative clustering (SLIC) algorithm is utilized to divides the infrared image into a background superpixel basic unit and a target superpixel basic unit and marks each basic unit as a background or target. Then, HTML 5 is used to reset the distribution of multiple identical small infrared target points to the original infrared image to prevent the extreme imbalance of the training data set from overfitting the model. Secondly, an adaptive enhanced SVM model is proposed to set the initial sample weight for the total number of samples in the training set. After that, samples are drawn to form a temporary training set to train the SVM classifier. Among them, in the weight iterative update stage, a higher misclassification cost is assigned to the wrong sample, so that the weight of the wrong sample increases faster and the number of algorithm iterations is reduced. In addition, the iterative process of the algorithm greatly optimizes the robust performance of a single classifier. Finally, we propose an adaptive threshold segmentation method to achieve the final target detection, so as to obtain accurate infrared target results. Extensive experiments in various complex scenarios proved that the proposed method is more robust and effective than 5 state-of-the-art contrast methods.