A scene recognition method in noisy environment

Abstract

The present invention relates to the technical field of a scene recognition method in a noisy environment, comprising the following steps: 1) using a sensor to acquire a scene image, which includes a marked sample; 2) separately analyzing the color image and depth image of the scene Perform feature extraction and feature expression, merge color image features and depth image features of the same group; 3) select feature selection algorithm to use the feature selection model obtained in the second step for the marked samples; 4) use classifier sort. The beneficial effect of the present invention is to accurately identify the scene in a noisy environment, ensuring that the sample also has a certain identification ability after being mixed with noise; thus improving the performance when the indoor scene data set is mixed with noise.

Description

technical field [0001] The invention belongs to the technical field of a scene recognition method, in particular to the technical field of a scene recognition method in a noisy environment. Background technique [0002] Generally speaking, scene classification can be regarded as a view-independent object recognition problem, and a scene is composed of a series of entities. For example, an indoor scene will contain chairs, tables, people, and bookshelves, and the placement of these things is not fixed. Accurate recognition of the scene helps to solve many practical applications, such as content-based image retrieval, robot path planning technology and image annotation, etc. Nowadays, scene recognition has attracted more and more attention from researchers. [0003] Extensive research has shown that the feature dimension of the image obtained after feature extraction is very high, limited by computing resources, and the high feature dimension will affect the scene recognitio...

AI Technical Summary

Problems solved by technology

[0003] Extensive research has shown that the feature dimension of the image obtained after feature extraction is very high, limited by computing resources, and the high feature dimension will affect the scene recognition system composed of RGB-D (combining color image and depth image) sensors practicality

Even though the existing feature selection technology can make high-dimensional features more concise and effective, the existing feature selection methods ignore the problem of a large amount of noise mixed in the sample. However, in practical applications, due to the complexity of the system The degree of accuracy and equipment processing accuracy are often mixed with a lot of noise, so the recognition effect of the existing feature selection algorithm will have certain limitations.

Images