Lightweight semantic segmentation method based on multi-scale visual feature extraction

Abstract

The invention discloses a lightweight semantic segmentation method based on multi-scale visual feature extraction, and the method comprises the following steps: network building: firstly building a lightweight convolutional neural network LitNet based on multi-scale feature extraction, extracting image features through a feature extractor, transmitting the features to a spatial pyramid module fused with hole convolution to extract multi-scale features of the image, and finally completing the feature integration through a simple up-sampling module to recover the image resolution; network training: employing a Tensorflow framework for building a network structure, employing a cross entropy function as a loss function, employing an Adam algorithm for optimizing training parameters, and adopting an early stop strategy in the training process to prevent network training overfitting so as to achieve the optimal training effect; network testing: inputting the test image into the network to obtain a semantic segmentation result, calculating mIoU and FPS, and evaluating the network performance. Through testing, the model size on the CamVid data set is 10M, the mIoU is 70.24%, the 34FPS can be reached, and the real-time segmentation requirement can be met.

Description

technical field [0001] The invention belongs to the technical field of image segmentation, and relates to a lightweight semantic segmentation method based on multi-scale visual feature extraction. Background technique [0002] In high-mobility autonomous decision-making terminal systems such as drones and unmanned vehicles, how to achieve accurate environmental perception is an important basis for system operation. Knowledge inference can be performed on the pictures collected by the equipment to complete the scene understanding of the equipment. Image semantic segmentation is an important branch in the field of AI and an important part of image understanding in machine vision technology. Semantic segmentation is a process from rough reasoning to fine reasoning, that is, by finding the category of image pixels, identifying the content and position in the picture, and finally completing the overall labeling of each object in the image to form an image mask or output The clas...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a lightweight semantic segmentation method based on multi-scale visual feature extraction, which solves the problem of large semantic segmentation network model and slow reasoning speed in the fields of existing semantic segmentation tasks

Images