The Study And Application Of Discrete Regularization Method For Detection Grassland

The grassland is a complex economic environment relating to the living things, climate, economy and other aspects. China has vast expanse of grassland resources, however, the rodents and hares have caused great damage to the grass. Our work depends on 863 plan project called "study on key digital technologies for managing the northeast grassland and herding industry in china in a resource efficient way" that is part of the state high-tech research and development plan program. We adopt dynamic way to detect the rat infestation and integrate with image segmentation technique, so that we can statistics and analyze the amount of rat and rat infestation situation. It is convenient to adopt an active response measures to reduce economic and ecological loss.Image segmentation is the key technique to image processing. Image segmentation bases on the image's features divides image into disjointed constituent regions or objects each of which has its homogenous feature. Those separated regions or objects are the objects of interest. Image segmentation is the key procedure from image processing to image analysis and has an important position. On the one hand, it is the foundation of object denotation and feature measure, on the other hand, object denotation, feature extraction and parameters measurement based on image segmentation are all abstract expression of the original images, which are convenient to image analysis and understanding.Image segmentation is widely used in pattern recognition and images retrieval, medical images processing. For the grass detection, the application background focuses on rat detection. We brought out a method called a digital detection method for rat hole and another method named a semi-supervised clustering of discrete regularization based on energy partition work on rat image. The first method includes thresh operator, multi-temp late mask, binary dilation operator and so on. We briefly introduced procedure and algorithms of the first method, analyzed the defaults and advantage combined with the experimental results. The first method depended on threshold, to overcome those problem, we decided to adopt the second method which is more effective and can automatic segment images.Discrete regularization method for grassland detection is composed of energy partition and discrete regularization methods and clustering operator. The energy partition is used to divide original image into block structure, get the same feature pixels together in the region and give region the mean value. Discrete regularization is used to remove the noise of image and smooth block structure, in order to reduce regions'contrast. The next step we adopt FCM or K-means clustering method to classify image with different class based on images'feature related to different objects. When background is complex, there are some errors appear in clustering result, so we need to fix the classified result. The clustering class as seeds, we integrate with discrete regularization semi-supervised classification method to correct FCM or K-means clustering. The discrete regularization semi-supervised classification method can remove singularity and statistic the number of regions without setting threshold. The background of this algorithm is detection rat, and the experiment shows that our algorithm is effective and we got ideal results. Compared to traditional method, our method can automatically detect the situation of rat, and the period is shorter and timely better. The discrete regularization method can also be applied to cell image, which is similar to rat images and is kind of multi-goal objects, and these objects belong to disconnect classification. But cell images have their own features, so we should use different processing steps. For cell images, we used K-means and discrete regularization operator, the experiment indicates that is an effective approach.