Application Research On Tephritid Fly Grooming Behavior Recognition Via Key Points Tracking And Spatio-Temporal Context

The grooming behavior of insects is an important part of their defense mechanism.Counting and analyzing the grooming behavior of insects,clarifying their types,laws and functions,can find more effective methods for pest control,which is conducive to the sustainable development of agriculture.The relevant literature shows that for the analysis and research of insect grooming behavior,most researchers at home and abroad manually record the start time and end time of each behavior by playing offline videos and observing with the naked eyes frame by frame.This method is time-consuming,labor-intensive,and error-prone.The data deviation increases with the increase of observation time,which affects the final research conclusion.In view of the above shortcomings of manual observation and the integration of cross-professional technologies,this paper proposes an innovative method for automatic recognition and statistical insect grooming behavior using computer vision and artificial intelligence technology.Combined with practical application scenarios,finally a simple and effective system for recognition and statistics of insect grooming behavior was designed.The system aims to significantly reduce the human and intellectual costs of plant protection related professionals in the research of insect grooming behavior,and provide researchers with reliable data for further research and analysis.The innovations of this paper are as follows: 1)Integrating artificial intelligence and plant protection majors to conduct research on insect dynamic behavior recognition.2)Combining deep learning methods with traditional image analysis methods,using the deep neural network(DNN)model to accurately extract the ROI,and then integrate the FFT and multi-frame gray momentum to extract the spatio-temporal features of the grooming behavior.3)A CNN model was trained to recognize and classify spatio-temporal features,which considered both accuracy and performance,and then an efficient recognition and statistics system for grooming behavior of Tephritid fly was designed.In this paper,taking the common crop pest Bactrocera minax as the example.The recognition system is used to classify and count the grooming behavior of insects,and these following work has been completed: 1)Insect behavior data collection and original data analysis,then established a database of insect grooming behavior features.2)Accurate extraction of insect behavior regions,and a high-performance spatio-temporal feature extraction algorithm for insect grooming behavior is proposed.3)A neural network model that can accurately and quickly classify the extracted spatio-temporal features is designed.4)Designed an end-to-end behavior recognition and statistics system,which has been applied in insect grooming behavior research.The proposed system aims to greatly reduce the human and intellectual cost of plant protection professionals in research on insect grooming behavior,and provide researchers with reliable data for further research and analysis.The experimental results show that the system not only greatly reduces the manpower,but also ensures the accuracy of insect behavior recognition and statistics.The system proposed in this paper is used to classify the grooming behavior of Bactrocera minax,the recognition accuracy is over 96%.Compared with the manual observation results,the coincidence rate is higher than 96%.This study focuses on the development theme of "smart agriculture",which provided a new analysis method for the research and statistics of related insects’ grooming behavior.Its application can improve the research efficiency of insect ecology research,and has good application prospects in plant protection,insect monitoring,pest control,and even animal behavior recognition research.