Forest unstructured scene segmentation method based on multispectral image fusion

Abstract

The invention discloses a forest unstructured scene segmentation method based on multispectral image fusion, and the method comprises the steps: introducing enhanced vegetation index (EVI) data, designing a parallel double-coding structure to extract the features of RGB and EVI for the conditions of complex background, multiple light rays, shadow interference and the like of a forest unstructured scene, and carrying out the segmentation of the forest unstructured scene. And multi-mode complementary features are formed in the encoding process. In addition, in the encoding stage, the receptive field of the network is increased by using expansion convolution, and the feature extraction process is optimized. And carrying out secondary fusion on the decoding features and fusion features generated by coding to obtain a multispectral fusion convolutional neural network, training the network, and inputting RGB and EVI images to realize semantic segmentation of the forest unstructured scene. The method effectively solves the problem that the unstructured segmentation method based on the RGB image is easy to have poor adaptability and wrong segmentation, and improves the accuracy and robustness of the semantic segmentation of the forest unstructured scene.

Description

technical field [0001] The invention belongs to the technical field of computer vision and unmanned vehicle environment perception, and relates to an unstructured scene segmentation method, in particular to a forest unstructured scene segmentation method based on multispectral image fusion. Background technique [0002] In recent years, autonomous driving technology has developed rapidly, and it has become a hot field for scholars from all walks of life to study and solve transportation problems. In the process of driving, providing comprehensive, accurate and reliable perception for autonomous vehicles is the premise of realizing safe and reliable autonomous driving. Among them, semantic segmentation technology has been widely studied as an important part of autonomous driving scene perception technology. At present, in the field of autonomous vehicle perception, the more mature visual semantic segmentation technology is mainly designed for structured road environments, su...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the above problems, the present invention discloses a forest unstructured scene segmentation method based on multispectral image fusion, which effectively solves the problem that most current segmentation methods based on RGB images are difficult to apply to complex unstructured scenes. Using different perception methods to construct feature descriptions of the same scene, further improving the accuracy and robustness of forest unstructured scene segmentation

Images