Digital Insect Identification Based On Support Vector Machine

The insect is huge in quantity, variety, the traditional identification method is lime consuming, and depends on the few insect classification experts, has long cycle, large workload, identification with the subjective bias and other defects. With the rapid development of computer science, automatic identification technology received widespread attention because of its advantages, such as large capacity, features, speed, good repeatability, reduce the subjective bias of investigators. Fast and automatic identification of insects based on automatic identification technology has high practical value, such as the rapid identification of the pest in agriculture is an important prerequisite for pest control work; in addition, it can provide a rapid identification method for the customs who lack of professional plant protection knowledge and the quarantine station, reduce the inflow of harmful insects. Insect classification model and feature extraction are two key problems in the development of high precision and the rapid method for automatic identification of insects, this paper studied the automatic insect identification algorithm from the two above aspects, results are reported as follows:In the development of automatic insect identification algorithm, choosing an appropriate classification model is one of the key issues. Template matching method is one of the early classics pattern recognition methods between different categories of simple similarity classification adopted by it, the distinction between capacity is not obvious; Principal component analysis(PCA) and kernel discriminant analysis(KDA) method based on statistics although to some extent improves the classification ability, but did not obtain satisfactory results; Artificial neural network(ANN) as a machine learning method, law of differences can be obtained by studying the insect, its identification speed, adaptability, and so has a greater increase, but its learning mechanism based on empirical risk minimization, is a linear modeling method, over learning, vulnerable to local minimum and other defects; Support vector machine (SVM) is a nonlinear machine learning method based on statistical learning theory, which is based on the structural risk minimization, can effectively prevent over learning, more adaptive to nonlinear characteristics of automatic identification of insects. Therefore in this study, SVM is used as the basic tool to build automatic identification model of insects.Insect image feature extraction is another key problem in automatic insect identification model. The effectiveness of extracted features directly determines the identification model recognition accuracy, ease their access to decide the identification of speed. Common image of such as color, texture, shape,can be more detailed characterization of insect images, but these characteristics are easily affected by image itself quality, light, and depends on the predicted effects of the image, under current technology and its practical application ability is limited. Mathematical morphology with mathematics method to describe and analyze the shape of the object, has the advantages of intuitive, efficient, easy to obtain, get extensive attention in the automatic insect identification field. Due to the sample heterogeneity, mathematical morphology feature extraction is not generally applicable to all categories, and too many features may also contain redundant features. So, it is necessary for feature selection, not only can simplify the model, can improve the identification accuracy of the model more.Based on the above method of automatic insect identification, first seven butterfly species to genus, belonging to Parnassiidae, Pieridae, constructed digital automatic identification model. Before using DrawWing software for seven species of butterfly wings internal vein automatically access point coordinates, and calculate the Euclidean distance between the adjacent intersections as mathematical morphological characteristics; then by the "one versus rest" method constructing binary classification model each type of sample with other samples; and then for each model based on SVM nonlinear feature selection, removing the useless or redundant features, and constructing the final classifier to retain characteristics, realizing the automatic identification of seven butterfly species.7prediction models of independent testing average accuracy up to98.64%.The automatic identification model for Hemiptera, Lepidoptera, Coleoptera,3orders and19families a total of automatic identification of34species of insects, extracting insect’s area and perimeter, transverse, longitudinal long, form factor, lobation, sphericity, circularity, eccentricity, roundness, hole number11mathematical characteristics, in level-order and total order section with two levels to achieve the automatic identification of34species of insects, have obtained better identification accuracy. The above two instances data shows the new method has good application prospect in the field of automatic insect identification.