Research on an Improved YOLOv7-based Intelligent Ant Nest Recognition Method for Dam Drone Images

ZHANG Shu; XU Yun-qian; LI Pei-cong; YUAN Ming-dao; ZHANG Xu-hui

doi:10.12396/znsd.2500307

","magId":"64451cbc-adb7-4dc7-a94f-a868087068d2","labelCn":"图1","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F001.jpg.ppt","labelEn":"Fig.1","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F001.jpg","titleEn":"Fig.1 Technology roadmap","titleCn":"图1 技术路线","id":"F1","captionEn":"Technology roadmap ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F001.jpg"},{"captionCn":"YOLOv7网络结构图 ","magId":"fa861127-7945-41a5-8115-5e5a388b7c87","labelCn":"图2","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F002.jpg.ppt","labelEn":"Fig.2","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F002.jpg","titleEn":"Fig.2 YOLOv7 network architecture diagram","titleCn":"图2 YOLOv7网络结构图","id":"F2","captionEn":"YOLOv7 network architecture diagram ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F002.jpg"},{"captionCn":"SENet注意力机制 ","magId":"b9843e2f-512b-443f-ba3e-84430a8d7be9","labelCn":"图3","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F003.jpg.ppt","labelEn":"Fig.3","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F003.jpg","titleEn":"Fig.3 SENet attention mechanism","titleCn":"图3 SENet注意力机制","id":"F3","captionEn":"SENet attention mechanism ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F003.jpg"},{"captionCn":"改进YOLOv7算法的主干特征提取网络 ","magId":"82cf9fd2-f633-4d50-9692-eefb905a6ca6","labelCn":"图4","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F004.jpg.ppt","labelEn":"Fig.4","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F004.jpg","titleEn":"Fig.4 Improving the Backbone Feature Extraction Network of YOLOv7 Algorithm","titleCn":"图4 改进YOLOv7算法的主干特征提取网络","id":"F4","captionEn":"Improving the Backbone Feature Extraction Network of YOLOv7 Algorithm ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F004.jpg"},{"captionCn":"无人机数据采集 ","magId":"4180126c-7536-4110-a845-f0b2c08ec06f","labelCn":"图5","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F005.jpg.ppt","labelEn":"Fig.5","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F005.jpg","titleEn":"Fig.5 Drone Data Collection","titleCn":"图5 无人机数据采集","id":"F5","captionEn":"Drone Data Collection ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F005.jpg"},{"captionCn":"数据增强效果图 ","magId":"0dad0597-8235-43d1-a8b4-1216b49c482e","labelCn":"图6","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F006.jpg.ppt","labelEn":"Fig.6","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F006.jpg","titleEn":"Fig.6 Data augmentation effect image","titleCn":"图6 数据增强效果图","id":"F6","captionEn":"Data augmentation effect image ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F006.jpg"},{"captionCn":"模型结果对比 ","magId":"2736cd3f-19ac-4675-85ab-133252c15bb9","labelCn":"图7","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F007.jpg.ppt","labelEn":"Fig.7","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F007.jpg","titleEn":"Fig.7 Comparison of model results","titleCn":"图7 模型结果对比","id":"F7","captionEn":"Comparison of model results ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F007.jpg"},{"captionCn":"原始YOLOv7算法检测结果 ","magId":"9dd46531-14da-49fe-aab3-5557a3778a7f","labelCn":"图8","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F008.jpg.ppt","labelEn":"Fig.8","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F008.jpg","titleEn":"Fig.8 Original YOLOv7 algorithm detection results","titleCn":"图8 原始YOLOv7算法检测结果","id":"F8","captionEn":"Original YOLOv7 algorithm detection results ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F008.jpg"},{"captionCn":"改进YOLOv7算法检测结果 ","magId":"60154203-c1df-4c81-b0d6-dd3ce03cfbc4","labelCn":"图9","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F009.jpg.ppt","labelEn":"Fig.9","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F009.jpg","titleEn":"Fig.9 Improve YOLOv7 algorithm detection results","titleCn":"图9 改进YOLOv7算法检测结果","id":"F9","captionEn":"Improve YOLOv7 algorithm detection results ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F009.jpg"},{"captionCn":"部分堤坝蚁穴可视化热力图 ","magId":"8574c3ea-0e58-4ed7-a640-6f7a6d509dc0","labelCn":"图10","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F010.jpg.ppt","labelEn":"Fig.10","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F010.jpg","titleEn":"Fig.10 Visualization heatmap of ant nests in dams","titleCn":"图10 部分堤坝蚁穴可视化热力图","id":"F10","captionEn":"Visualization heatmap of ant nests in dams ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F010.jpg"},{"captionCn":"堤坝蚁穴智能识别系统 ","magId":"8d307cce-7efa-478b-a001-6e4c0840bc58","labelCn":"图11","pptUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F011.jpg.ppt","labelEn":"Fig.11","figUrl":"1007-2284-2026--1-190/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F011.jpg","titleEn":"Fig.11 Intelligent recognition system for ant nests in dams","titleCn":"图11 堤坝蚁穴智能识别系统","id":"F11","captionEn":"Intelligent recognition system for ant nests in dams ","thumbnailUrl":"1007-2284-2026--1-190/thumbnail/DA2C639F-751F-475f-A962-6F88C1DA6B3E-F011.jpg"}],"article":{"keywordList_cn":["水利工程","无人机巡查","YOLO v7","蚁穴识别","注意力机制"],"juan":"0","zhaiyao_cn":"

针对水利工程蚁穴隐患人工巡查效率低的问题，提出了一种基于改进YOLOv7的无人机图像智能检测方法。通过在YOLOv7主干网络的3个特征层融合SENet通道注意力机制，强化模型对蚁穴目标的特征提取能力，构建包含1445张堤坝蚁穴无人机图像的增强数据集用于模型训练。实验结果表明：改进模型在精确率、召回率和mAP指标上分别达到91.2%、95.4%和91.5%，较原始模型提升2.7%、3.2%和3.4%，有效解决了蚁穴图像几何特征弱导致的漏检问题。基于PyQt5开发的智能识别系统实现了蚁穴智能识别模型工程化应用，支持图像、视频输入与多维度结果展示，为堤坝蚁害智能巡检提供了有效解决方案。

","endNoteUrl_en":"https://irrigate.whu.edu.cn/EN/article/getTxtFile.do?fileType=EndNote&id=21705","bibtexUrl_cn":"//www.fangxunbeng.com/irrigate/CN/article/getTxtFile.do?fileType=BibTeX&id=21705","articleType":"A","abstractUrl_en":"//www.fangxunbeng.com/irrigate/EN/10.12396/znsd.2500307","qi":"1","id":21705,"nian":2026,"bianHao":"1769492733265-25575523","juanUrl_en":"https://irrigate.whu.edu.cn/EN/Y2026","shouCiFaBuRiQi":"2026-01-27","qiShiYe":"190","accepted":"2025-04-26","received":"2025-03-05","qiUrl_cn":"https://irrigate.whu.edu.cn/CN/Y2026/V0/I1","lanMu_cn":"水利工程","pdfSize":"1668KB","risUrl_cn":"//www.fangxunbeng.com/irrigate/CN/article/getTxtFile.do?fileType=Ris&id=21705","title_cn":"基于改进YOLOv7的堤坝无人机图像蚁穴智能识别方法研究","doi":"10.12396/znsd.2500307","jieShuYe":"195","keywordList_en":["hydraulic engineering","drone surveillance","YOLO v7","ant nest recognition","attention mechanism"],"endNoteUrl_cn":"//www.fangxunbeng.com/irrigate/CN/article/getTxtFile.do?fileType=EndNote&id=21705","zhaiyao_en":"

To address the inefficiency and low accuracy of manual inspections for ant nest hazards in hydraulic engineering， this study proposes an intelligent detection method for drone images based on an improved YOLOv7 model. By integrating SENet channel attention mechanisms into three feature layers of the YOLOv7 backbone network， the model’s capability to extract features of ant nest targets is enhanced. An augmented dataset containing 1，445 drone images of embankment ant nests was constructed for model training. Experimental results demonstrate that the improved model achieves precision， recall， and mAP values of 91.2%、95.4% and 91.5%， respectively， representing improvements of 2.7%、3.2% and 3.4% over the original model. This effectively mitigates missed detection issues caused by weak geometric features in ant nest images. An intelligent recognition system developed with PyQt5 enables the engineering application of the model， supporting image and video input alongside multi-dimensional result visualization， thereby providing an effective solution for intelligent inspection of embankment ant hazards.

","bibtexUrl_en":"https://irrigate.whu.edu.cn/EN/article/getTxtFile.do?fileType=BibTeX&id=21705","abstractUrl_cn":"https://irrigate.whu.edu.cn/CN/10.12396/znsd.2500307","juanUrl_cn":"https://irrigate.whu.edu.cn/CN/Y2026","qiUrl_en":"//www.fangxunbeng.com/irrigate/EN/Y2026/V0/I1","risUrl_en":"https://irrigate.whu.edu.cn/EN/article/getTxtFile.do?fileType=Ris&id=21705","title_en":"Research on an Improved YOLOv7-based Intelligent Ant Nest Recognition Method for Dam Drone Images","hasPdf":"true"},"authorList_en":[{"deceased":false,"xref":"¹^,²","name_cn":"张舒","email":"gdskyzs@163.com","xref_en":"¹^,²","name_en":"ZHANG Shu"},{"deceased":false,"xref":"¹^,²","name_cn":"徐云乾","xref_en":"¹^,²","name_en":"XU Yun-qian"},{"deceased":false,"xref":"¹^,²","name_cn":"李培聪","xref_en":"¹^,²","name_en":"LI Pei-cong"},{"deceased":false,"xref":"¹^,²","name_cn":"袁明道","xref_en":"¹^,²","name_en":"YUAN Ming-dao"},{"deceased":false,"xref":"¹^,²","name_cn":"张旭辉","xref_en":"¹^,²","name_en":"ZHANG Xu-hui"}]}; new mag_vue({ el: '#metaVue', data:window.metaData, loading:true, doTextFun:function(text){ return doDataJsonText(text); }, doAfterFun:function(selector,jsonData){doMagVueAfter(selector,jsonData);} }); var pars = {attachType:'PDF',id:$("#articleId").val(),json:true}; jQuery.post($("#mag-common-contextPath").val()+"/EN/article/showArticleFile.do?" + new Date().getTime(), pars, function(data){ var json=null; if(data.match(/^\[json\].*/g)){ json=eval("("+data.replace(/^\[json\]/g,"")+")"); } if(json && json.status == 1){ //有权限 }else{ //无权限 $("#no-access-div").show(); } }); mag_ajax({ url:mag_currentQikanUrl() + '/EN/article/getRichHtmlJson.do?articleId='+$("#articleId").val(), dataType:"json", async:"false", success:function(json){ if(json.article){ new mag_vue({ el: '#bodyVue', data:json, doTextFun:function(text){ return doDataJsonText(text); }, doAfterFun:function(selector,jsonData){doMagVueAfter(selector,jsonData);} }); } } }); //img .mag_rich_graphic var _nlmdtdXml = $("#article_nlmdtdXml").val(); var magDir = _nlmdtdXml.replace(/\/[^\/]+$/,''); $("img.mag_rich_graphic").each(function(){ $(this).attr("src",mag_currentQikanUrl()+"/"+magDir+"/"+$(this).attr("src")); }); //video .mag_shipin $("video.mag_shipin").each(function(){ $(this).attr("src",mag_currentQikanUrl()+"/"+magDir+"/"+$(this).attr("src")); if($(this).attr("poster") && $.trim($(this).attr("poster"))!=''){ $(this).attr("poster",mag_currentQikanUrl()+"/"+magDir+"/"+$(this).attr("poster")); } }); //a .mag_body_media_a $("a.mag_body_media_a").each(function(){ $(this).attr("href",mag_currentQikanUrl()+"/"+magDir+"/"+$(this).attr("href")); }); //去掉正文中的中文文字“多媒体文件” $("div.mag_body_media_file b").remove(); //altmetric if($.trim($("#article_doi").val())!=''){ $.ajax({ url:'https://api.altmetric.com/v1/doi/'+$.trim($("#article_doi").val())+'?callback=altmetric_callback', //url:'https://api.altmetric.com/v1/doi/10.1007/s13238-016-0264-7?callback=altmetric_callback', dataType:'jsonp' }); } mag_rich_menu(jQuery); $(function(){ $('#autoMenu').autoMenu(); }) $("#autoMenu").niceScroll({ cursorcolor: "#ccc", //滚动条的颜色 cursoropacitymax: 0, //滚动条的透明度，从0-1 touchbehavior: false, //使光标拖动滚动像在台式电脑触摸设备 true滚动条拖动不可用 cursorwidth: "6px", //滚动条的宽度单位默认px cursorborder: "0", // 游标边框css定义 cursorborderradius: "3px", //滚动条两头的圆角 autohidemode: true, //是否隐藏滚动条 true的时候默认不显示滚动条，当鼠标经过的时候显示滚动条 zindex: "auto", //给滚动条设置z-index值 railvalign:'defaul', railpadding: { top: 0, right: -6, left: 0, bottom: 0 }, //滚动条的位置 }); window.autoMenuFlag=true; $("#relatedArticles").niceScroll({ cursorcolor: "#ccc", //滚动条的颜色 cursoropacitymax: 0, //滚动条的透明度，从0-1 touchbehavior: false, //使光标拖动滚动像在台式电脑触摸设备 true滚动条拖动不可用 cursorwidth: "6px", //滚动条的宽度单位默认px cursorborder: "0", // 游标边框css定义 cursorborderradius: "3px", //滚动条两头的圆角 autohidemode: true, //是否隐藏滚动条 true的时候默认不显示滚动条，当鼠标经过的时候显示滚动条 zindex: "auto", //给滚动条设置z-index值 railvalign:'defaul', railpadding: { top: 0, right: -6, left: 0, bottom: 0 }, //滚动条的位置 }); $('body').click(function(){ $('.daochu_div').hide(); $('.qipao').hide(); }) $('.qipao').click(function(event){ event.stopPropagation(); }) /*参考文献显示隐藏*/ $('.biaotijiantoush1').click(function(event){ if($(this).attr('class')=='glyphicon glyphicon-triangle-top biaotijiantoush biaotijiantoush1'){ $(this).parents('.mag_main_zhengwen_left_div_ckwx_table_benwenyiny').next('.mag_main_zhengwen_left_div_ckwx_table_ckwx_xiangqing').stop().slideUp(); $(this).attr('class','glyphicon glyphicon-triangle-bottom biaotijiantoush biaotijiantoush1'); }else if($(this).attr('class')=='glyphicon glyphicon-triangle-bottom biaotijiantoush biaotijiantoush1'){ $(this).parents('.mag_main_zhengwen_left_div_ckwx_table_benwenyiny').next('.mag_main_zhengwen_left_div_ckwx_table_ckwx_xiangqing').stop().slideDown(); $(this).attr('class','glyphicon glyphicon-triangle-top biaotijiantoush biaotijiantoush1'); } event.stopPropagation(); }); //加载补充材料 mag_ajax_update({ ele_id:'showArticle_suppls', url:mag_currentQikanUrl() + '/EN/article/showArticleSuppls.do', data:{articleId:$("#articleId").val()} }); //加载专题 mag_ajax_update({ ele_id:'showArticle_subjects', url:mag_currentQikanUrl() + '/EN/subject/showSubjectListByArticle.do', data:{articleId:$("#articleId").val(),forwardJsp:'showArticle'} }); //渲染公式 mag_mathJax(); }); function altmetric_callback(json){ if(json && json.score){ if(json.score - parseInt(json.score) > 0){ $("#altmetric-score").html(parseInt(json.score)+1); }else{ $("#altmetric-score").html(parseInt(json.score)); } $(".main_content_center_right_duoweidu_div").css("width","25%"); $("#altmetric-a").show(); $("#altmetric-a").attr("href",json.details_url); } } function richClickAuthor(name){ var searchSQL = "((("+mag_advSearchValReplace(name)+"[Author]) AND 1J[Journal]) AND year[Order])"; mag_request({url:mag_currentQikanUrl() + '/EN/searchresult',data:{searchSQL:searchSQL}}); } function richClickKeyword(keyword){ var searchSQL = "((("+mag_advSearchValReplace(keyword)+"[Keyword]) AND 1J[Journal]) AND year[Order])"; mag_request({url:mag_currentQikanUrl() + '/EN/searchresult',data:{searchSQL:searchSQL}}); } function mag_share(){ $("#bShare_more_a").click(); } function loadRelatedArticles(){ if($("#showRelatedArticles") && $("#showRelatedArticles").length>0){ //加载过了 }else{ mag_loading(); mag_ajax_update({ ele_id:'relatedArticles-content', url:mag_currentQikanUrl() + '/EN/article/relatedArticle.do', data:{id:$("#articleId").val(), forwardJsp:'showArticle'}, complete:function(){ mag_layer_close(); } }); } } function showMetrics(){ mag_window_location(mag_currentQikanUrl() + '/EN/metrics/'+$("#articleId").val()); } function piaofuFn(a,eleId){ var content=''+$("//www.fangxunbeng.com/irrigate/EN/10.12396/#"+eleId).prop("innerHTML")+'

'; piaofu(a,content); } function huoqukuanduimg(){ // } function doDataJsonText(text){ var str = text; //var _nlmdtdXml = $("#article_nlmdtdXml").val(); //var magDir = _nlmdtdXml.replace(/\/[^\/]+$/,''); return str; } function doMagVueAfter(selector,jsonData){ var _nlmdtdXml = $("#article_nlmdtdXml").val(); var magDir = _nlmdtdXml.replace(/\/[^\/]+$/,''); //处理fig $(selector + " .mag-rich-fig[magid]").each(function(){ var magId = $(this).attr("magid"); for(var j=0;j"; ss += "

"; ss += ""; var shuomingTitle = ""; var shuomingContent = ''; if(fig.titleEn && fig.titleEn!=''){ shuomingTitle = fig.titleEn; if(fig.titleContentEn && fig.titleContentEn!=''){ shuomingContent += fig.titleContentEn; } } if(shuomingTitle==''){ if(fig.titleCn && fig.titleCn!=''){ shuomingTitle = fig.titleCn; } if(fig.titleContentCn && fig.titleContentCn!=''){ shuomingContent += fig.titleContentCn; } }else{ if(fig.titleCn && fig.titleCn!=''){ shuomingContent += "

"+fig.titleCn+"

"; } if(fig.titleContentCn && fig.titleContentCn!=''){ shuomingContent += fig.titleContentCn; } } ss += '

'+shuomingTitle+'

'; if(shuomingContent!=''){ ss += shuomingContent; } ss += '

'; ss += '

Full size|PPT slide

'; $(this).html(ss); break; } } }); //处理表格 $(selector + " .mag_main_table[magid]").each(function(){ var magId = $(this).attr("magid"); for(var i=0;i"; } if(tab.titleContentCn && tab.titleContentCn!=''){ shuomingContent += tab.titleContentCn; } } ss += '