引用本文:凌敏,胡华浪,曾爱萍,陶双华,王亚鑫.基于无监督和YOLOX的农村房屋建设状态识别算法[J].中国农业信息,2022,34(3):52-60
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 46次   下载 48 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于无监督和YOLOX的农村房屋建设状态识别算法
凌敏1,胡华浪2,曾爱萍1,陶双华2,王亚鑫2
1.西南交通大学中国土地信息大数据研究院,四川成都 610097;2.农业农村部大数据发展中心,北京 100020
摘要:
【目的】 为了提升自然场景中农村在建房屋的识别准确率,并为后续的农村违建房屋智能化监管提供技术支撑。【方法】 文章基于无监督聚类和YOLOX目标检测算法,发展了一种乡村房屋在建状态识别方法。首先,构建在建房屋无监督聚类模型,并以此对在建房屋进行类别精细划分,使得不同类别之间特征差异较大,相同类别特征差异较小,其次,再使用划分好的类别制作房屋检测数据集,并训练YOLOX目标检测模型对在建房屋进行识别,最后,在在建房屋数据集上设计模型对比实验,以此验证算法有效性。【结果】 实验结果表明:在在建房屋识别任务中,基于无监督聚类和YOLOX的在建房屋识别算法mAP为83.27%,比采用原始数据(不进行在建房屋类别划分)训练的YOLOX算法mAP提升了7.91%,同时比采用人工划分类别的YOLOX算法mAP提升了5.08%。【结论】 因此该文方法有效提升了乡村房屋在建状态的识别精度,同时也为具有复杂场景和多个不同状态的目标进行识别时,提升识别准确率提供一种有效且可靠的解决思路。
关键词:  无监督聚类;YOLOX;乡村房屋;目标检测;状态复杂
DOI:10.12105/j.issn.1672-0423.20220306
分类号:
基金项目:四川省自然资源科研项目“自然资源要素时空生态演化链构建与智能检索技术研究” (Kj-2022-(2))
Rural housing construction status recognition algorithm based on unsupervised and YOLOX
Ling Min1, Hu Hualang2, Zeng Aiping1, Tao Shuanghua2, Wang Yaxin2
1.Chinese Research Institute of Land and Big Data,Southwest jiaotong University,Chengdu Sichuan 610097,China;2.Big Data Development Center,Minstry of Agriculture and Rural Affairs of the People’s Republic of China,Beijing 100020,China
Abstract:
[Purpose] In order to improve the identification accuracy of rural houses under construction in natural scenes,and provide technical support for the subsequent intelligent supervision of rural houses under construction.[Method] Based on unsupervised clustering and YOLOX target detection algorithm,a rural house status recognition method is developed in this paper,First,the unsupervised clustering model of houses under construction is constructed,and the houses under construction are divided into categories,so that the feature differences between different categories are large and the feature differences of the same category are small. Secondly,the divided categories are used to make the housing detection data set,and the YOLOX target detection model is trained to identify the houses under construction,Finally,a model comparison experiment is designed on the housing data set under construction to verify the effectiveness of the algorithm.[Result] The experimental results show that:in the task of identifying houses under construction,mAP of the house under construction identification algorithm based on unsupervised clustering and YOLOX is 83.27%,compared with the YOLOX algorithm trained by raw data(no classification of houses under construction),its mAP improved by 5.08% and 7.91% compared with the YOLOX algorithm based on manual classification.[Conclusion] Therefore,this method can effectively improve the identification accuracy of rural houses under construction,It also provides an effective and reliable solution to improve the recognition accuracy of data with complex scenes and multiple different states.
Key words:  unsupervised clustering;YOLOX;rural housing;target detection;state complexity