摘要
针对DeeplabV3+网络在土地分类中存在的分割不完全、边缘细节丢失等问题,以岳城水库附近居民区为研究区,提出一种基于改进的DeeplabV3+无人机影像土地利用分类模型。首先,获取研究区无人机影像数据并建立相应数据集;其次,引入轻量级网络MobileNetV2代替DeeplabV3+的主干特征提取网络,大幅度降低模型参数量,提高模型计算速度;最后,加入CA注意力机制,减少细节丢失,提高分割精度。实验结果表明,改进模型不仅在分割性能方面优于原始DeeplabV3+模型,有效解决了道路断连和分割不完全等问题,提高了地物的分割精度,在分割效率方面也有很大提高。
Aiming at the problems of incomplete segmentation and missing edge details of DeeplabV3+ network in land classification,a land use classification model based on improved DeeplabV3+ unmanned aerial vehicle(UAV)image was proposed,taking the residential area near Yueceng Reservoir as the study area. Firstly,obtain the UAV image data in the study area and establish the corresponding data set;Secondly,the lightweight network MobilenetV2 is introduced to replace the backbone feature extraction network of DeeplabV3+,which greatly reduces the parameters of the model and improves the calculation speed of the model;Finally,Coordinate Attention(CA)mechanism is added to reduce the loss of detail and improve the segmentation accuracy. The experimental results showed that the improved model not only has better segmentation performance than the original DeeplabV3+ model,but also effectively solves the problems of road disconnection and incomplete segmentation,improves the segmentation accuracy of ground objects,and greatly improves the segmentation efficiency.
作者
刘粉粉
王贺封
张安兵
李家驹
马鹏飞
LIU Fen-fen;WANG He-feng;ZHANG An-bing;LI Jia-ju;MA Peng-fei(College of Earth Science and Engineering,Hebei University of Engineering;College of Mining and Geomatics Engineering,Hebei University of Engineering;Handan Key Laboratory of Natural Resources Spatial Information,Handan 056038,China)
出处
《软件导刊》
2022年第11期130-136,共7页
Software Guide
基金
国家自然科学基金(42071246)
河北省自然基金委重点项目(D2021402007)
河北省自然科学基金项目(E2020402006)。
