Research On Camouflage Human Target Detection Based On Deep Learning

With the rapid development of computer vision,object detection technology has penetrated into all aspects of social life.As one of the core problems in the field of computer vision,human object detection in complex environment has received more and more widespread attention,and has broad application prospects in military reconnaissance,personnel search and rescue,intelligent driving and video surveillance.At present,camouflage human target detection has become one of the difficulties in the field of complex environment human target detection,because military camouflage targets are usually in a complex environment,overlapping occlusion is more common,the target is often integrated with the surrounding environment background,with very high concealment,very low recognition characteristics,resulting in the search and positioning of human targets in the process of false detection and missed detection.Therefore,the use of artificial intelligence technology to quickly and accurately detect and identify military camouflage personnel is of great significance,this paper applies deep learning technology to complex environment camouflage human target detection tasks,mainly carried out the following four aspects of work:1.The military camouflage personnel dataset MCPD was constructed and produced.Taking the military camouflage video collected on the Internet in the complex environment in the field as the original material,the high-definition image containing camouflage human target is intercepted and screened by video framing,and the human target in the image is manually annotated,so as to successfully construct a multi-scene,multi-directional and high-quality military camouflage personnel dataset,which lays a data foundation for the training and verification of the detection method in this paper.2.A camouflage human target detection method based on attention mechanism is proposed.Camouflage human targets are usually located in a complex environment,overlapping occlusion is more common,resulting in the algorithm is difficult to extract the rich feature information of military camouflage targets,and with the deepening of the number of network layers,the loss of feature information is more serious,which in turn affects the detection accuracy.To solve the above problems,a camouflage human object detection algorithm TC-YOLOv5 s based on attention mechanism is proposed,and firstly,the self-attention module is embedded at the end of the feature extraction and feature fusion network on the basis of the YOLOv5 s framework,which strengthens the extraction of global information of the image,models the dependence between all pixels,and enhances the recognition ability of the algorithm for camouflage personnel and environmental background.Then,a convolutional attention module is added to the feature fusion network,which further strengthens the ability of the algorithm network to extract the features of camouflaged personnel,weakens the attention to the surrounding environment,and effectively improves the anti-background interference ability of the algorithm.3.A camouflage human target detection method based on multispectral frequency is proposed.In the camouflage human target detection task in a multi-battlefield environment,there is a phenomenon that the target is highly similar to the environmental background,and the degree of discrimination is low,resulting in false detection and missed detection,and the traditional attention module is easy to cause the lack of feature information because it only focuses on a single frequency.To solve the above problems,a camouflage human target detection algorithm based on multispectral frequency MBM-YOLOv5 s is proposed.Firstly,YOLOv5 s is selected as the basic framework,and a multispectral channel attention module is embedded in the backbone feature extraction network to enhance the information propagation between the target features of the camouflaged human body and improve the recognition of the target and complex background by the network.Secondly,the original feature pyramid network in the feature fusion network is replaced with a weighted bidirectional feature pyramid network to achieve efficient bidirectional cross-scale connection and weighted feature fusion.Finally,the Mixup data enhancement strategy is used to simulate overlapping occlusion scenarios,strengthen the learning ability of the network model on complex samples,and improve the algorithm to effectively improve the detection accuracy of camouflaged human targets.4.A military camouflage personnel detection system based on deep learning was designed and built.According to the requirements of real scenarios such as intelligent combat,military reconnaissance guidance and natural disaster personnel search and rescue,the visual interface design and module function are realized by tools such as Pycharm and Py Qt5,and then the military camouflage personnel detection system based on deep learning is successfully designed by carrying the MBM-YOLOv5 s network model with good experimental training before,and the whole process of system development is introduced in detail,and finally the accuracy and reliability of the system for the detection and identification of military camouflage personnel in multibattlefield environments are tested.In summary,this paper applies deep learning technology to the task of camouflage human target detection in complex environments,and effectively realizes the accurate detection and recognition of military camouflage personnel in multi-battlefield environment through algorithm optimization and system detection design.