On-line Identification And Counting Of Sex-pheromones Lured Orchard Pest Based On Machine Vision

The monitoring of pest species and population in orchard is the basis of orchard pest control and management. Traditional pest monitoring is mainly relied on manual investigation and statistics, which cost much labor and time. An on-line orchard pest identification and counting method based on machine vision was studied, in addition, the prototype of monitoring equipment and system were designed. This work can promote the application of automatic recognition and classification technology based on image vision in the orchard pest monitoring. Main research includes the following four parts:(1) An image segmentation algorithm based on shape factor and separation point location was presented. In this method, a shape factor was used to justify whether the connected region was one touching region or not. Local segmentation point and separation points were determined by contour tracking and stripping layer by layer. Finally, the touching object was separated by connecting the local point and the two boundary separation points successfully. The results showed that the average segmentation error rate and the average segmentation efficiency rate of the proposed segmentation method were7%and92.65%respectively, in the yellow peach moth (Conogethes punctiferalis (Guenee)) experiment, and the average segmentation error rate and the average segmentation efficiency rate of the proposed segmentation method were2.24%and97.8%, repectively, in the oriental fruit moth (Grapholitha molesta (Busck)) experiment.(2) Two feature extraction methods ware discussed based on ideas of morphology invariant feature and morphology irrelevant feature, respectively. In the image Zernike moment method, low order moment and high order moment were built and compared with each other in the orchard pest identification. In the color texture combination method, six feature combinations were constructed based on different color spaces and different texture features.(3) Orchard pest classification method was studied based on multi-class support vector machine. The performances of three parameter optimization algorithms (Grid Search, Genetic Algorithm and Particle Swarm Optimization) were compared and the best one was genetic algorithm (GA). The classification results for test samples showed that the high accuracy rate was derived from the texture feature group, wavelet decomposition of H, S and V channel images based on’db4’wavelet, whose accuracy was100%with the lowest time cost.(4) On-line orchard pest monitoring equipment and identification algorithm were designed and verified. The hardware structure and software function were implemented according to the parameters in the orchard pest monitoring application. Numerous experiments were conducted and the results showed that the average recognition rate was above95%in the single kind of pest experiments with non-thouching and slight thouing, and the average recognition rate was84.2%in the severe touching experiments. In addition, in the identification test for multi pest species, the threshold value should be determined according to the pest size, and its average recognition accuracy was87.67%.