Stored Grain Pest Detection Method Based On Image Processing

Our country is in the front rank in grain production, storage preservation and consumption every year. Annual total stored grain is about 500 billion kg. During the period of storage, stored grain suffered a great losses account for about 0.2% because of getting mildew, stored-insects and rats. The losses caused by stored-insects account for half of the total losses. Thus, to detect the stored-insects precisely is the key to decrease the loss of the stored grain. And it's also the scientific basis for control decision-making to the stored-insects. In recent years, domestic and foreign researchers detected the stored-insects outside the grains by traps, samples and machine vision etc; And they also detected the stored-insects inside the grains by conductivity detection, near-infrared reflectance (NIR), X-ray imaging detection etc. But these methods can't provides us stored-insects' varieties and density information accurately and real time. Applying image process techniques to detect stored-insects possesses following advantages:cheap in price, high efficiency, high accuracy, pollution-free, small quantities of work. This paper applied image process methods to enhance, to denoise, to segment and to detect the edge of stored-insects images. And it laid the base for characteristic abstraction and accurate recognition.The main research work in this paper is as follows:Firstly, in the process of image collection and transmit, stored-insects image may be polluted by noise which caused by shocking, dots, etc,and it may reflect the image quality and subsequent process. This paper applied the methods of mean filtering, median filtering and wiener filtering and the wavelet analysis methods to denoise the stored-insect images. In the process of stored-insect image collection, the stored-insect image may bring about variety lack fidelities which may lead to misunderstanding. This paper applied the methods of histogram equalization, grey adjustment and power transformation three methods for space domain image enhancement and Wavelet analysis methods to enhance the stored-insect images for the frequency domain image.Secondly, the effective segmentation of stored grain insect images is the basis of extraction of stored insects feature and accurate recognition. This paper firstly applied the methods of histogram based threshold segmentation method, Otsu method the method base on level set to segment the stored grain insect images with quite different foreground and background; and then applied the image segmentation methods based on graph cuts theory to segment the stored grain insect images with similar foreground and background.Thirdly, the effective detection of stored grain insect images is the important basis of the feature extraction and choice of stored pests. For grey stored-insects image edge detection, this paper firstly applied the classical edge detection methods as Robert algorithm, Prewitt algorithm and Canny algorithm to detect the edges of the stored grain insect images; and then the results of the stored-insect image edge detected by the methods above has been compared. Owing to the binary image possesses the advantages as high processing efficiency and timeliness, etc, it holds the important space in stored-insects image real-time processing. Thus the four methods above are applied to detect the edge of the binary image of stored insects containing noises. After the comparison of the experimental results, it concluded that the edge detection method base on grey theory has the benefits of edge locating accurately, edge detected continuously and effectively and the better performance in de-noiseing for binary image.