Research On Self-Organizing Feature Map Neural Network Algorithm In Image Recognition

Images, as human being's most important information source, are intuitive and easily understood. With the development of science and technology, especially the development of computer and digital image technology, many new theories, approaches, algorithms, methods, and techniques come into being. And they are put into practice in respects of scientific research, agricultural monitoring, and industrial production, etc. They make huge contributions to social development and are conducive to improve people's living standard. The relevant management of precision agriculture particularly greenhouse agriculture requires the effective collection and extraction of information on crop growth. By using the image information, we can control the crop's growth conditions, collection of fruit, removal of weeds and the spraying of all kinds of nutrients.Based on the shortage of the classification methods in the weed segmentation, the extra-green character segmentation and Self-Organizing Feature Map neural network were integrated to develop a G-SOFM space classification model to classify the weed picture. The method is that using two feature vectors of the gray of excessive green and the normalized, after the processed of extra-green character segmentation. The results show that by using G-SOFM space classification model classifying better than extra-green character segmentation method 20%. After the algorithm, the image denoising method is used, and then the recognition will rate up to 94%.The paper also points out that after we get access to the features of images, we carry out spatial transformation, and then make a feature clustering analysis on the transformed space. The algorithm improves the clustering effect. Lastly we take a feasibility test in the embedded platform. It turns out that the algorithm is feasible.