High-resolution image rape extraction method based on deep learning algorithm

Abstract

The invention provides a high-resolution image rape extraction method based on a deep learning algorithm, the model fuses DRN and an improved U-Net semantic segmentation model, comprises two core modules of Encoder and Decoder, and realizes autonomous learning of spatial features and spectral features of a high-resolution No.2 image and automatic extraction of a rape pixel level. According to the model, a Gaofen-2 image in the flowering period (from the end of February to the middle of April) of oilseed rape is taken as a data source, training and testing are performed through marked oilseed rape samples, the oilseed rape samples are increased through a data enhancement algorithm, and an optimized oilseed rape extraction model is obtained by adjusting model parameters. When the model is applied, a data input cutting strategy and a result output cutting strategy are provided to realize rape extraction of input images of any size. Compared with a traditional classification analysis method based on pixel and statistical theory and an object-oriented classification algorithm, the method makes full use of the spatial and spectral features of the Gaofen-2 image, realizes autonomous learning of semantic features for rape extraction, and improves the precision of rape extraction and the model generalization ability.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and more specifically relates to a high-resolution image rapeseed extraction method based on a deep learning algorithm. Background technique [0002] Rapeseed is the first of the five oilseed crops in my country (including rapeseed, soybean, peanut, sunflower and sesame), and is the only winter oilseed crop. It is of great significance to the adjustment of industrial structure, food security, national economy and social development. [0003] China's rapeseed production is affected by the geographical environment and farmers' planting behavior, showing the characteristics of fragmented planting. Conventional survey methods are time-consuming and laborious, and it is difficult to meet the normalized monitoring needs of large-scale spatial distribution of rapeseed; and medium and low spatial resolution remote sensing It is also difficult for images to meet the resolution requ...

AI Technical Summary

Problems solved by technology

[0003] China's rapeseed production is affected by the geographical environment and farmers' planting behavior, showing the characteristics of fragmented planting. Conventional survey methods are time-consuming and laborious, and it is difficult to meet the normalized monitoring needs of large-scale spatial distribution of rapeseed; and medium and low spatial resolution remote sensing It is also difficult for images to meet the resolution requirements for the extraction of fragmented rapeseed; the successful launch of the Gaofen-2 satellite provided basic data support for the rapid and accurate monitoring of the spatial distribution of crops and area estimation, but the "same object, different spectrum" and " The ubiquity of the phenomenon of "different objects with the same spectrum" limits the extraction accuracy of the traditional classification analysis method based on pixel and statistical theory. Although the object-oriented classification algorithm can effectively improve the classification accuracy of remote sensing images, it uses texture and shape and other spaces. The features are limited and have obvious regional differences, which limits the promotion and generalization ability of a single model

Images