Research Of Image Recognition About Drosophila’s Courtship Behavior

With the continuous development of image processing, pattern recognition, artificial intelligence and other related technologies, the recognition of insect’s behavior was becoming a research focus. That study, which was based on theory about image processing and pattern recognition, selected drosophila in courtship period and realized automatic identification of drosophila courtship behavior by drawing on related research from home and abroad. Firstly image of drosophila was drawn from the video files,and then the characteristic parameters,which reflecting its courtship state,was extracted base on image segmentation.Finally,the BP neural network was used for identifying the drosophila’s courtship behavior.Additionally,a method was also proposed to improve the recognition accuracy,based on target tracking technology.The main research contents are as follows:(1) By contrast,the DirectShow was chosen to extract images from video, and gray-scale images was used for subsequent segmentation operation;At the same time, the image preprocessing methods was proposed based on background generation and updation,according to the characteristics of the drosophila images.(2)The automatical segmentation method of drosophila image was proposed by utilization of Otsu, mathematical morphology and histogram. And a part of drosophila,which were not mating, were screened advancely,through the study of the positionial relationship and the mse of difference image about mating drosophila.(3)Through comparative analysis, the first four low-order moments of Hu-moments, drosophila contour’s area, perimeter, and roundness were selected as the feature parameter for recognizing drosophila’s courtship behavior.(4)The BP neural network,trained by levenberg-marquardt algorithm, was selected to classify drosophila. According to the experimental results, the structure of the network was adjust to 7-14-1, and the threshold of the output value was set as 0.5. The average recognition rate was 87%,which could demonstrate the effectiveness of the neural network.(5)The drosophila’s tracking algorithm, based on neighborhood trajectory prediction, profile similarity and Kalman filter was proposed. And the algorithm’s effection for reducing the error rate and misings rate was also discussed.(6) The recognition system was achieved with the use of Emgu CV,Visaul Studio 2010 and C Sharp.The system was simple to operate and could be used for recognizing drosophila’s courtship behavior automatically and achieving artificial comparison.The above-mentioned research results could be theoretical basis and technical support for the automated identification of drosophila’s courtship behavior, while saving a lot of manpower and ensuring the recognition accuracy rating. So it had important theoretical and practical value, and was helpful for conducting large-scale experiments.