建设项目用地预审是自然资源管理部门对建设项目是否符合法律法规和土地利用总体规划的初步审查，在用地预审环节，往往存在未批先建、边批边建等违法用地行为，对于违法行为的有效监测是一项重要任务。提出一种基于深度学习多模型融合的用地预审项目疑似动工监测方法，通过构建建设用地预审项目业务模型、利用国土变更调查及多源卫星遥感影像制作专题解译样本、训练多专题深度学习模型，并搭建多模型融合环境、构建自动化分析计算模型，实现用地预审自动化监测。实验结果表明，该方法能够有效识别用地预审项目范围内疑似动工图斑，综合识别率优于71.3%，效率高、人工干预少，为建设项目用地预审土地利用现状监测提供有效手段。

The pre-examination of construction project land-use is a preliminary examination conducted by the natural resource management department to determine whether the construction project complies with laws, regulations, and the overall land-use overall plan. In the pre-examination process of land-use, there are often illegal land-use behaviors such as building before approval or building while approval is being carried out.Effective monitoring of illegal behaviors is an important task. We proposed a method for monitoring suspected construction of land-use pre-examination projects based on deep learning multi-model fusion. By constructing a business model for construction land-use pre-examination projects, using territorial change survey and multi-source satellite remote sensing images to produce thematic interpretation samples, training multi-thematic deep learning models, building a multi-model fusion environment, we constructed an automated analysis and calculation model to achieve automated monitoring of land-use pre-examination. The experimental results show that this method can effectively identify suspected construction site spots within the scope of land-use pre-examination projects, with a comprehensive recognition rate of better than 71.3%, and has high efficiency, minimal manual intervention, which can provide an effective means for monitoring land-use pre-examination for construction projects.