Identification And Classification Study Based On The Images Of Aphid Natural Enemies

Prediction and control of pests in the field make an impact which can not be ignored on national food security and agricultural development. There are several methods for the Pest Control and Management according to the pests in the field, such as chemical control, biological control, physical control, agricultural control and genetic control. The chemical control brings up negative effect to agricultural development, since the method kills the pest as well as the natural enemies and causes some problems, such as pesticide residues, environment pollution, and destroying of the ecological balance. The biological control will be the method has more widespread application, because of the unique control target, no pollution, no residual, being safe to lives and being friendly to environment. The investigation on the pest and their natural enemies should come first for prediction and control of agricultural pests by getting the density of the pest and their enemies in the field. For aphid natural enemies, the traditional methods of investigation and identification have the lack of inconvenient count and higher error.In recent years, advanced information technologies are applied constantly in the Pest Control and Management aspect by a number of domestic and foreign research institutions and researchers, with the development of artificial intelligence, image identification, computer vision and other high-tech. It is of important theoretical significance and practical value in the aspect of Pest Control and Management, knowing the information of the natural enemies by catching their images with real-time monitoring system, a digital camera or other digital products, and getting their types and number with image processing and computer vision techniques timely and more accurately.In this paper, the classification and identification study based on three main types of aphid natural enemies has been discussed, by extracting global features of the images of the natural enemies, such as texture, shape, color, as well as local features with the descriptors SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Features); and classifying three types of aphid natural enemies with the neural network classification and similarity matching methods, the main contents and results are as follows: (1) By extracting seven global features about the images on three main types of aphid natural enemies including color, shape and texture aspects, such as, color histogram, color moments, shape factor, rectangular degrees, gray co-occurrence matrix, gray level smoothing matrix and Tamura texture features, this paper gives classification and identification effect individually based on each feature, after analyzing the feature vector.(2) The classification is studied by integrating all the global features extracted with the neural network for identification. After the discussions about the classification effects of OSU-SVM and LS-SVM, it is found that the SVM identification method using LS-SVM toolbox can achieve better identification effects.(3) It extracts local features of the images on three main types of aphid natural enemies with the two descriptors SIFT and SURF.(4) The extracted local features of three main types of aphid natural enemies are matched by using nearest neighbor matching method based on similarity measurement, and the experiment results show that the method base on SURF descriptor is more efficient, faster and well robust.(5) The identification method based on SURF with histogram intersection can improve the identification percent of aphid natural enemies. The method based on SURF and LBP is experimented, and the result show that the identification percent of the flies increased.