A Study Of Diptera Insects Grooming Behavior Recognition By Integrating Multi-objective Tracking And Spatio-temporal Feature Model

Insect grooming behavior is common and habitual for many insects,usually using their feet to scrub different parts of their bodies with each other,generally to keep their bodies clean and to disperse group odors and pheromones.In-depth studies on multiobjective tracking of insects and recognition of their grooming behavior can help professionals systematically analyze the physiological,neurological,and pharmacological basis of their grooming behavior and thus develop safe and efficient pest control strategies.At present,the recording and quantification of grooming behavior of insects are mainly recorded manually,specifically by professionals observing video footage of grooming behavior frame by frame,so as to achieve the purpose of recording the type and duration of grooming behavior of insects,which is time-consuming and laborious and prone to errors due to human eye fatigue.Secondly,there are also a small number of machine vision recognition and recoding methods for insect grooming behavior,and they have also achieved good results,but these methods are only for single-object insects as research objects,which do not meet the actual insect behavior recording requirements.To address the above problems,this paper proposes an innovative method of multiobject insect grooming behavior recognition based on computer vision and artificial intelligence technology with interdisciplinary integration,using which we can automatically recognize and classify the grooming behavior of insects efficiently and accurately.In this paper,we take the dipteran insect Bactrocera minax as an example to recognize and classify its grooming behavior,and the main work is as follows:(1)Collecting grooming behavior dataset and establishing behavior feature library,collecting a total of 190 minutes of video of Bactrocera minax multi-object behavior dataset and a total of 8 categories of behavior category feature library;(2)Extracting the behavior occurrence domain accurately in real time,and each Bactrocera minax individual in the dataset accurately tracked and detected;(3)Extracted spatio-temporal feature information of insect grooming behaviors,and efficiently and clearly generated spatio-temporal feature information maps of each grooming behavior;(4)Designed a convolutional neural network to accurately recognize and classify insect grooming behaviors.The innovations of this paper include:(1)A non-invasive,pure machine vision method for insect grooming behavior recognition is proposed,which is a crossapplication research of artificial intelligence + agriculture and provides new ideas and methods for the research of computer vision technology in the field of insect behavior recognition.(2)The existing Deep Sort object tracking algorithm is improved,and the detector part uses the SE attention module added to the Yolov5 s model instead of the native Faster R-CNN model,and the tracker part introduces an acceleration model instead of the native constant velocity model,which improves the accuracy of insect object tracking;(3)The research explores a variety of spatio-temporal feature learning and classification models,and finally designs a new Res Net network that incorporates the SE module—SE-Res Net new model,which can effectively and accurately recognize and classify insect grooming behaviors.Using the method proposed in this paper,the grooming behaviors of a total of 23 Bactrocera minax were recognized and recorded in four test videos,and the results showed that the average accuracy exceeded 96%,while the standard deviation was less than 3%,comparing with existing methods,the method in this paper has a higher accuracy and the deviation was controlled within a certain range.Therefore,the method in this paper significantly improves the efficiency while ensuring the accuracy of grooming behavior recognition.Its application can improve the efficiency in the field of insect ecology research and can have a broad development prospect in the fields of plant protection,pest monitoring,insect behavior analysis,and even animal behavior recognition.