Based On The Bag Of Words Model Of Insects Image Classification System Research

With the rapid developments of information technology and the widely applications ofthe artificial intelligence (AI), the people began to use the computer vision theory for theresearch of the insects image classification system. It takes advantage of the mass storage andhigh speed of the computes to make the insect recognition tasks automate for the need of thespeed and the accuracy in the industry and the agriculture areas.In the insect recognition field, the global features including the color and the shape etc.have been used for the representation of images. However, these approaches are not very goodin the local areas of image description and these features may be un-robust with insects sameinsects image background, illumination, scale, rotation, translation of the factors. Therefore, itis necessary to adopt the new features operator to extract information in the images of theinsects and avoid the redundant information on insect classification results of the existence ofthe influence.SIFT descriptor is invariant to the scale, translation, rotation and affine, which is theessence of the content of the image characteristic, can be in observation conditions change isbig, keep out, mixed and disorderly the complex conditions of interference to image contentto describe.This paper adopts SIFT descriptor instead of global characteristics, where bag ofwords model (BoW) is introduced for the representation of the image features.The descriptionof the insect recognition has an important roles in the semantic understanding, which alsoavoids the redundant information to affect the results of classification.In addition, this paper uses AdaBoost algorithm for building classifier. We giveAdaBoost algorithm including its detailed analysis of the main ideas and the scope ofapplication and propose a system function flow chart for the development. The experimentsresults with7kinds of617with complex background image of insects show that the BoWmodel is simple and effective for the insect recognitions.