Research And Application Of Lightweight YOLOv4 Algorithm Based On Mixed Domain Attention And Feature Cross Fusion

Rapid and accurate detection of potential risks and hidden dangers of community electric bicycles,prevention of fire hazards,or timely detection and alarm of areas where fires have occurred,is one of the core issues of community fire protection.Focusing on this issue,this article takes YOLOv4 as the research foundation and conducts algorithm improvement research from three aspects: proposing a YOLOM algorithm based on mixed depthwise convolutional kernels to solve the problem of excessively large algorithm parameters,and proposing a YOLOMC algorithm based on feature cross fusion to solve the problem of low fusion efficiency,A YOLOMCS algorithm based on separated mixed domain attention was proposed to solve the problem of channel information communication barriers and feature extraction efficiency caused by group convolution.The effectiveness and feasibility of the proposed algorithm were verified through experiments.At the same time,the proposed algorithm was applied to practical application scenarios and a community electric bicycle safety hazard detection system was designed and implemented.The main research work and achievements include:1.A YOLOM target detection algorithm based on mixed depth separable convolution is proposed.In view of the fact that the computing power of the device is limited,and the lightweight detection algorithm is more available.According to the low cost computing characteristics of the current CPU,based on the YOLOv4 algorithm framework,the problem of the large number of parameters of the traditional YOLOv4 algorithm and the slow detection speed on the CPU device is solved by introducing the mixed depth separable convolution.The experimental results show that the algorithm can reduce the algorithm parameters and improve the detection speed,while maintaining high detection accuracy.2.A YOLOMC target detection algorithm based on feature cross fusion is proposed.Aiming at the problem of semantic information loss caused by traditional fusion methods in the process of feature fusion and the reduction of target detection accuracy,the YOLOM algorithm is used as the framework to assign weight to semantic features,and a weight self-learning method is designed to achieve feature cross-fusion.Experimental results show that this algorithm has higher detection accuracy and fewer algorithm parameters than traditional single feature fusion.3.A YOLOMCS target detection algorithm based on separated mixed domain attention is proposed.In order to improve the efficiency of semantic information processing,key information is selected from many information and the problem of channel information communication gap is alleviated.A separate mixed domain attention mechanism is designed to make the network model connect with global semantic information,pay attention to the semantic information required by the task,and strengthen the information flow between channels.Taking YOLOMC algorithm as the framework,the experimental results show that compared with other attention mechanisms,this method can detect target information quickly and effectively,and further improve the detection accuracy.4.A community electric bicycle safety hazard detection system was developed using Py Qt and My SQL technology.We have designed and implemented functions such as algorithm selection,real-time monitoring,target detection,abnormal alarm,We Chat reception,and result display.Abnormal alarm reception through We Chat can improve the efficiency of dangerous situation handling and reduce the cost of repeated app development.We have successfully applied the algorithm proposed in this article to practical application scenarios.