Semantic segmentation method based on multi-scale convolutional neural networks (CNNs)

Abstract

The embodiment of the invention discloses a semantic segmentation method based on multi-scale convolutional neural networks (CNNs). The method includes: acquiring intra-modal features in high-resolution aerial-images and LiDAR (Light Detection And Ranging) point cloud data; carrying out inter-modal feature extraction and classification on the basis of the multi-scale convolutional neural networks;and using a multi-scale segmentation method to extract ground object boundaries, eliminate salt and pepper effects, and optimize classification results. By implementing the example of the invention,a method of combining the multi-scale CNNs and multi-scale segmentation (MRS) post-processing is used for semantic segmentation of the high-resolution aerial-images and the LiDAR point cloud data.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a semantic segmentation method based on a multi-scale convolutional neural network. Background technique [0002] Deep learning is a new field in machine learning research. Its purpose is to establish a neural network that simulates the human brain for analysis and learning. It imitates the mechanism of the human brain to interpret data, such as images, sounds and texts. This concept originates from the research of artificial neural network, which can be understood as the development of neural network. Its essence is to learn more useful features by building a machine learning model with multiple hidden layers and massive training data, thereby improving classification or prediction. accuracy. Commonly used deep learning models include convolutional neural networks (CNN), deep belief networks (DBNs), and so on. [0003] Convolutional neural network is a kind of artificial neu...

AI Technical Summary

Problems solved by technology

[0008] 3) Although the encoder-decoder CNN architecture can upsample the low-resolution features derived from the pooling layer to the input resolution, because the upsampling layer reconstructs the appearance of the object instead of the shape, resulting in the upsampled object boundary become blurred and irreversible

However, most of the existing LiDAR point cloud data and high-resolution aerial image fusion methods are loose fusion of intra-modal features, ignoring the statistical characteristics between modalities.

In addition, the convolutional neural network (CNN) uses tiles as input data, and the pixels at the edge of the tiles are prone to salt and pepper effects, resulting in uncertain labeling results

Images