Semi-automatic vegetation remote sensing sample selection method based on convolutional neural network

Abstract

The invention provides a semi-automatic vegetation remote sensing sample selection method based on a convolutional neural network. The method comprises the following steps of S1, constructing a multi-source multi-temporal remote sensing database, S2, dividing vegetation in the research area into m vegetation groups, S3, constructing a deep residual network, and changing the number of network output categories into m, S4, using the public vegetation sample data set to train a deep residual network, and enabling model parameters to be suitable for vegetation group classification, S5, segmenting the image of the research area into vegetation group pattern spots by adopting a multi-scale segmentation algorithm, S6, selecting n representative samples of m vegetation types, and taking m*n samples as seed samples of sample selection, S7, calculating network output of all individual pattern spots in the research area, and S8, calculating network output of the seed sample. The problem of sample scarcity in mountainous area vegetation classification is solved to a certain extent.

Description

technical field [0001] The invention relates to the technical field of mountainous vegetation remote sensing, in particular to a method for selecting semi-automatic vegetation remote sensing samples based on a convolutional neural network. Background technique [0002] Vegetation map reflects the quality, quantity, type and spatial distribution of vegetation resources, and is a key tool for fully, rationally and sustainably utilizing vegetation resources. At present, there are two main methods of vegetation mapping: traditional vegetation mapping based on field surveys and remote sensing vegetation mapping supported by remote sensing data. The compilation of traditional vegetation type maps is mainly based on a large number of ground surveys, combined with the clear drawing of some aerial photos in the room. Although this method has a high accuracy rate, it is time-consuming and labor-intensive, and is often only able to cover a small area due to factors such as terrain, en...

AI Technical Summary

Problems solved by technology

In order to solve the problem that traditional vegetation mapping consumes a lot of manpower, material resources and time, and the training samples are scarce in vegetation remote sensing classification, the present invention designs a semi-automatic vegetation sample selection technology based on deep convolutional neural network. This technology makes full use of historical sampling Data and other prior knowledge, construct a set of universal knowledge transfer model, improve the automation of sample collection in the target research area, and solve the problem of scarcity of samples in mountainous vegetation classification to a certain extent

The invention can solve the problem of scarcity of samples for remote sensing classification of mountain vegetation to a certain extent, and greatly improves the automation of vegetation mapping

Images