Method for interpreting high-resolution remote sensing image by using combined loss function HRnet

Abstract

The invention provides a method for interpreting a high-resolution remote sensing image by using a combined loss function HRnet. The method comprises the following steps of high-resolution remote sensing image land utilization classification data set-semantic separation network-classification result-result evaluation. The deep learning algorithm is used for high-resolution remote sensing image land utilization classification, on one hand, manpower and material resources input in the manual visual interpretation process can be reduced to a great extent, and a computer can be used for rapidly and efficiently carrying out the interpretation task of the high-resolution remote sensing image; and subjective behaviors in a manual feature extraction process can be avoided to a great extent, and the classification precision is improved.

Description

technical field [0001] The invention relates to the field of land use classification, in particular to a method for interpreting high-resolution remote sensing images using a combined loss function HRnet. Background technique [0002] The land use classification of high-resolution remote sensing images is the basic problem in the field of land cover extraction. Because the land use classification standard is affected by both natural and social factors, there is a great mutual interference in the information of different land types. How to further improve the land use The classification accuracy exploited is an issue faced in current research. Manual interpretation is the most commonly used method with high accuracy. It depends on the size, shape, color and tone, shadow, position, texture, resolution, topography and other characteristics of the ground objects, and divides the image map into different landforms. However, this method requires the interpreter to have rich exper...

AI Technical Summary

Problems solved by technology

However, when extracting features, downsampling to 32 times the original image results in serious loss of detail information, and the obtained segmentation results are relatively rough.

The U-net model proposed by Ronneberger et al. in 2015 is an extension of the FCN network. The main improvement is the introduction of a symmetrical encoder-decoder (Encoder-Decoder) structure. The encoder (Encoder) is used to extract features, and the decoder (Decoder) uses To restore the original image, the segmentation accuracy has been improved, which has been recognized by scholars and has been applied and improved accordingly. However, the model loses the spatial detail position information during the feature extraction process, and the segmentation accuracy is limited.

When the land types are complex, the boundaries are not clearly defined, and land types of different sizes are to be extracted, the segmentation results of the semantic segmentation network using the cross-entropy loss function have low accuracy, and the phenomenon of misclassification and omission is serious, which can no longer meet the requirements of high-resolution remote sensing images. The land classification task of

In the existing semantic segmentation network, in the process of extracting features, the resolution of the feature map is usually reduced first and then restored, and rich semantic information is obtained, but the spatial detail information in the image is lost, resulting in inaccurate accuracy of small-sized ground objects. is extracted

Images