Moving Target Detection Algorithm Based on SIFT Feature Matching

Domain-specific image classification based on improved residual networks

“…Aimed at the problems of changing shooting angles and many occluders of military images, Xie Zeqi et al [7] was considering that the Fourier Descriptors has the invariance of translation and rotation, firstly used the Canny algorithm of contour extraction to extract the contour features of the image, and then performed Fourier transform on the image. Using the obtained Fourier transform as dataset, and then combined with the weighted SRC method to achieve image classification; Zhang Chunlei et al [8] firstly used edge detection operators Marr-Hildreth and Sobel to process the data set respectively, and then sent the processed image to the network to achieve feature extraction, and then standardized fused feature extraction results obtained by the two operators, and finally sent to the Softmax classifier to achieve image classification; Based on the selective search algorithm, Song Yiming et al [9] were using LBP and HOG algorithm to extract feature, then fused the acquired shallow features (LBP) and deep features (HOG) to break the information limitation brought by only single-layer features, and the support vector machine method was used to process the fused features to achieve classification; Zhang Shanwen et al [10] used the improved SRILBP and PHOG to achieve feature extraction for military targets, and then used the improved CCA to fuse the extracted features to obtain more abundant features information. Finally sent the processed result to K-nearest neighbor classifier and implemented classification.…”