Research On Image Object Detection Algorithm Based On Improved YOLOv4

As a technology that identifies the object category and location in images or videos,object detection plays an important role in safety monitoring,automatic driving,medical image processing and other fields.It is difficult for traditional object detection methods to meet the requirements of detection accuracy and speed in practical applications.In recent years,object detection methods based on deep learning have developed rapidly.The research is conducted on the YOLOv4 method and the specific work is as follows :1.Three lightweight convolutional network models named as Mv2-DSC-YOLOv4,Mv3-DSC-YOLOv4 and G-DSC-YOLOv4 respectively are designed to solve the problems of high amount of parameters,complex model structure and the difficulty in deploying on mobile or embedded devices for YOLOv4 method.The proposed models extract feature information based on different lightweight networks,and the depthwise separable convolution is utilized to make the network structure simpler.The experimental results show that the Mv3-DSC-YOLOv4 method achieves the best detection effect among the three proposed methods.Compared with the YOLOv4 method,the detection speed is increased by 59.73 frames per second,and the parameters and model size are reduced by 52.36 M and 199.72 MB,respectively.2.A novel method named as real-time object detection method based on channel attention mechanism and multi-spatial pyramid pooling is proposed to avoid the disadvantages of an enhancement to the representational power of the deep feature maps of the feature fusion network for the SPP module,higher computational complexity and the difficulty in highlighting important channel features for the feature maps of the detection head network in YOLOv4 method.Since multiple receptive fields are fused after extracting multi-scale information by multi-spatial pyramid pooling,the characterization ability of the shallow,middle and deep feature maps is strengthened for the feature fusion network.By utilizing the squeeze-and-excitation channel attention mechanism to model interdependencies between channels,the weight of each channel is adaptively recalibrated to make the network pay more attention to important features.Moreover,the depthwise separable convolution is exploited to reduce the parameters of the feature fusion and detection head networks.The experimental results show that the mean average precision of the proposed method is higher than that of the state-of-the-art methods,while the average speed of the method reaches 33.70 FPS,which meets the real-time requirements.Compared with YOLOv4,the parameters and model size are reduced by 27.85 M and106.25 MB,respectively.The presented method not only improves the detection accuracy,but also reduces the computational complexity compared to the baseline method.