Image classification method based on HOG features and DMD

Abstract

The invention relates to an image classification method based on HOG features and a DMD, and belongs to the technical field of image classification. The image classification method comprises the steps: extracting the HOG features from a significant color matrix obtained by carrying out color-based dynamic mode decomposition on an image, obtaining similar pictures after the features are extracted, then carrying out model training on the similar pictures to obtain a new weight vector and a separation distance, and finally classifying pictures to be classified. According to the method, the recognition accuracy of small sample images and single sample images is high; the recognition accuracy of images with complex backgrounds is high; the time complexity and the space complexity of the algorithm are low; according to the method, a neural network result selection problem and a local minimum value problem are avoided; and the algorithm has good generalization for high-dimensional and nonlinear classification problems.

Description

technical field [0001] The invention relates to an image classification method based on HOG features and DMD, belonging to the technical field of image classification. Background technique [0002] The Histogram of Oriented Gradient (HOG) feature is a feature descriptor used for object detection in computer vision and image processing. It forms features by calculating and counting the gradient direction histogram of the local area of ​​the image. Compared with other feature extraction methods, HOG features have good geometric invariance and optical invariance, and are widely used in the field of image recognition. [0003] Dynamic mode decomposition (DMD) is a data-driven approach that does not require an accurate decomposition of highly complex systems into their respective coherent spatiotemporal structural equations, but uses coherent structures that grow, decay, and oscillate over time to solve or approximate the system. The coherent structure is also called DMD mode....

AI Technical Summary

Problems solved by technology

Compared with the KNN classifier, the SVM classifier is more suitable for processing high-dimensional data, but due to the limitations of its own algorithm, it takes a lot of time to process large batches of image data, and there is no uniform standard for kernel function selection. Judging based on experience, which leads to unsatisfactory accuracy and time complexity of the algorithm

The BPNN algorithm has a strong nonlinear mapping ability and has a high degree of self-learning and self-adaptive ability. The CNN algorithm has a greater advantage in processing high-dimensional data. Both of them have high image classification accuracy, but both have relatively long training time. Long, the training results are difficult to converge to the global minimum and there are shortcomings such as local minimization problems, which lead to high time complexity and computational complexity in the field of image classification

[0011] The purpose of the present invention is to be committed to solving the technical defects of the above-mentioned algorithm in terms of time complexity and algorithm complexity, while ensuring a higher classification accuracy

Images