Remote sensing image ground object identification method and system, and computer readable storage medium

Abstract

The invention provides a remote sensing image ground object identification method and system, and a computer readable storage medium. The method comprises the steps of collecting an original sample remote sensing image for training; performing data enhancement processing on the collected original sample remote sensing image to obtain an enhanced sample remote sensing image; constructing a multi-scale dense convolutional network; training the multi-scale dense convolutional network in combination with the original sample remote sensing image and the enhanced sample remote sensing image; and after the multi-scale dense convolutional network is trained, identifying a ground object in a remote sensing image to be identified through the multi-scale dense convolutional network, and marking the identified ground object. The densified hierarchical connection mode of the DenseNet network is adopted to reconstruct the network, the network depth is deepened, and transmission of feature information is enhanced; and a multi-scale feature conversion layer is constructed, and image feature information extracted by the convolution kernel is enriched, so that the network achieves the purpose of relatively high accuracy of remote sensing image ground object identification.

Description

technical field [0001] The present invention relates to the technical field of computer vision, in particular to a remote sensing image feature recognition method, system and computer-readable storage medium. Background technique [0002] Object recognition in remote sensing images refers to the process of linking each pixel in an image to a feature category label. This process is image semantic segmentation, but it can be thought of as pixel-level image classification. Semantic segmentation is a typical computer vision problem that involves taking as input some raw data (e.g., planar images) and converting them into masks with highlighted regions of interest, where each pixel in the image is defined according to its The belonging object of interest is assigned a class ID. Object recognition in remote sensing images is a key technology in geographic information systems, and it plays a very important role in many fields such as land planning, disaster prevention, unmanned a...

AI Technical Summary

Problems solved by technology

The implementation principles of these methods are different, but they basically use the low-level semantics of the image, including information such as the color, texture and shape of the image pixel, but the actual segmentation effect is not ideal when encountering complex scenes.

[0004] At present, deep learning is mainly used for related research, including the introduction of a full convolutional network (FCN) with an encoding-decoding structure, but FCN has problems such as loss of semantic information and lack of research on the correlation between pixels; while using unpooling The operating SegNet ensures the integrity of high-frequency information, but ignores the information between the neighbors of pixels in the lower-resolution feature map; and the DeconvNet with a fully connected layer has a larger network model and is more difficult to train

Images