Research Andapplication Of Compressive Sensing In The Estimation Of Broadband Signal Source Number

It is most basic work that recognition and classification of plants in plant research, andextremely significant for the division of floristic species, the exploration of floristicevolution and the relationship of floristic. Also, it has important application value inidentifying better varieties in modern agriculture. So it is the most simple and effective wayto identify and classify plant, using the shape and texture of plants.In this paper, we selected the500images of the leaves of French phoenix, trident maple,pears and other10kinds of blade as the research object of this article, which were chosenfrom the plant leaf image database (ICL) in the intelligent computing laboratory of Chineseacademy of sciences in He Fei. Firstly we proceed the chosen images by the gray-scaleprocessing, the binarization processing, morphological erosion to filter and remove holes andpetiole. Then, we can extract the shape and texture feature for the pretreated image. Throughcalculating the area and perimeter, using the image contour, convex hull and minimumbounding box, we can extract the5kinds of shape feature about rectangle rate, elongationrate, round rate, density rate and invariant moments and Gray level co-occurrence matrixmethod was used to extract5kinds of texture feature about entropy, energy, homogeneity,contrast and correlation. We select40images for each plants, and use400imagescharacteristic parameters as the training sample, and then choose one kind of plant leaves ofcharacteristic parameters as the test sample to look for its ownership in the training sample,where the method of sparse representation is used. Because the training samples arecomposed by multiple characteristic parameters, so we can determine its ownership in thesample by the adaptive weight the sparse representation.Finally, we conducted the experiments and simulation for the methods in this paper, andcompared with the other method, which based on neural network and SVM classification.The results show that the identification rate of the two methods above were89.6%and91.4%respectively. In this paper, the recognition rate of two methods reached94.6%and94.6%respectively, thus we can prove the recognition method on sparse representation has the betterclassification effect.