Design And Implementation Of Pigeon Cross Maze Cognitive Training And Behavior Recognition System

Pigeon is an important animal model for studying brain cognitive mechanism because of its high cognitive ability.Maze experiments are widely used to study the cognitive abilities of animals.Researchers analyze the target cognitive abilities through the behavior of animals in maze experiments.In traditional research,animal behavior information is mainly recorded by artificial methods,which will cause interference to experimental animals and lead to inaccurate measurement results.In recent years,researchers have proposed a video-based animal behavior recognition method,which reduces the interference to animals,but how to design an appropriate animal behavior recognition method is the key problem.Therefore,it is of great significance to design pigeons’ cross maze cognitive training system and construct pigeons’ behavior recognition algorithm for quantitative evaluation of pigeons’cognitive ability.This paper designs and builds the hardware and software of pigeon cross maze cognitive training system,and tests the cognitive training function of the system.The pigeon target detection algorithm,the codec model of pigeon key point detection and the ConvLSTM network model of pigeon behavior recognition were constructed.The experimental data of pigeon behavior in the training system were analyzed,and the recognition of pigeon cognitive behavior in the cross maze was realized.The main research work and results of this paper include:(1)A cross maze training system for pigeon target recognition is designed.According to the specific requirements of pigeon target cognition experiment,the hardware scheme and training process of the training system are designed.The hardware part mainly realizes the control of each module by PLC,and then realizes the automatic training of pigeon cognition.It mainly includes four modules:target stimulus module,behavior feedback module,reward module and video capture module.(2)The improved YOLOv3 algorithm model for pigeon detection is designed to realize the accurate detection of pigeons in the cross maze.The improved YOLOv3 algorithm model is designed for pigeon detection in the cross maze;.Through experimental tests,the average accuracy mAP of the algorithm model designed in this paper for pigeon detection is 90.77%,which is 2.57%higher than the YOLOv3 algorithm and meets the real-time requirements.Finally,the pigeons’ cognitive ability to target was tested by comparing the number of frames detected before and after training in the area where the target appeared.(3)A coding and decoding model is designed to detect the key points of pigeons,and the three key points of the head,back and tail of pigeons in the cross maze are detected accurately.According to the demand of pigeon keypoint detection,the codec model of pigeon keypoint detection was designed,and the information acquisition between different key points was further expanded through the empty convolution.Through experimental tests,the model in this paper can accurately identify the key points in the head,back and tail of pigeons,and the average accuracy of pigeon key point detection is 90%,higher than that of other algorithms.Finally,based on the identification of the key points on the back of the pigeons,the trajectory map corresponding to a video during the target cognition training of the pigeons in the cross maze was drawn.(4)A ConvLSTM network model was designed for pigeon behavior recognition,which realized the accurate recognition of four behaviors of pigeons in the cross maze:static,straight walking,turning around and eating.According to the requirement of pigeon behavior recognition,ConvLSTM network model was designed.Firstly,the optimal time window length of pigeon behavior recognition is obtained through experiments.Under the optimal time window length,ConvLSTM network was used to test and compare with other behavior recognition methods.The results show that the average accuracy of the proposed model in pigeon behavior recognition is 94.45%,which is higher than the average accuracy of other algorithms.The effect of the whole training system was verified by the number of different behaviors of pigeons before and after training.