Texture recognition method based on local binary threshold learning network

Abstract

The invention relates to a texture recognition method based on a local binary threshold learning network. The method comprises steps: 1, a to-be-classified texture image data set D is prepared, and the data set D is divided to a training data set Dt and a test data set Dv; 2, the local binary threshold learning network is built, the training data set Dt is inputted, and the local binary threshold learning network is trained through error sensitive term reverse propagation and a random gradient algorithm, wherein the local binary threshold learning network comprises one input layer, one threshold coding layer, two convolution layers, three down sampling layers, one full connection layer and one output layer; and 3, the test data set Dv is inputted to the well-trained local binary threshold learning network to verify a training result. According to the texture image classification method based on the local binary threshold learning network, through learning the structure information of the texture features, the method is applicable to texture image recognition in a small sample condition.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a method for classifying texture images using local binary networks based on threshold value learning. Background technique [0002] Texture analysis is an active research topic in the field of computer vision, and has long played an important role in many fields such as object recognition, remote sensing analysis, content-based image retrieval, etc. Texture features contain a series of important information such as the structure distribution of the object surface and its relationship with the surrounding environment. It is of great significance for the research and application of computer images, especially for the classification of images. Effective classification is often inseparable from Open the description of the image texture. [0003] Traditional image classification methods usually include three steps: feature extraction, feature expression, and cla...

AI Technical Summary

Problems solved by technology

However, hundreds of layers of networks and tens of millions of data cannot be separated from the support of expensive hardware devices while ensuring the actual effect, and will also consume a lot of computing time

More importantly, in the face of actual application requirements, it is often impossible to obtain enough training samples or sufficient hardware equipment to ensure time cost

In such cases, feature learning often loses its advantages

Images