云南师范大学学报（自然科学版）2025，Vol.45Issue(6)：21-25,5.DOI:10.7699/j.ynnu.ns-2025-065
基于改进随机森林的视频监控场景感知识别方法
Video Surveillance Scene Perception and Recognition Method Based on Improved Random Forest
摘要
Abstract
In the process of perception and recognition of video surveillance scenes,the basic random forest algorithm can only rely on fixed splitting criteria to complete scene classification,and its ability to learn hierarchical abstract features hidden in video data is insufficient,resulting in a low accuracy(ACC)of the final perception result.Therefore,based on the improvement of random forest,a new video surveillance scene perception recognition method was designed.First,the Backbone-ResNet50 and feature weighting mechanism were employed to analyze video surveillance images and extract key feature information.Then,an adaptive splitting criterion was introduced to replace the traditional ran-dom forest splitting mode,enhancing the feature learning capability of the random forest and construc-ting an intelligent perception model.Finally,multi-modal perception data was collected for different scenarios,and specific targets within the scene were identified using multi-modal feature information.The experimental results showed that the ACC of the scene perception obtained by this method re-mained above 95％,achieving high-quality processing of monitoring data.关键词
改进随机森林/视频监控场景/多模态数据/特征加权/自适应分裂准则/感知识别Key words
Improved random forest/Video surveillance scenarios/Multimodal data/Feature weighting/Adaptive splitting criteria/Perception recognition分类
信息技术与安全科学引用本文复制引用
江运峰,赵春泽,刘真,李文力,应卓君..基于改进随机森林的视频监控场景感知识别方法[J].云南师范大学学报（自然科学版）,2025,45(6):21-25,5.基金项目
成都市科技局科技资助项目(2024-YF05-02417-SN). （2024-YF05-02417-SN）
