COMPARATIVE ANALYSIS OF IMAGE RECOGNITION METHODS OBTAINED FROM SENSORS OF THE VISIBLE SPECTRUM

Abstract

Today, in an era of integration of artificial intelligence systems in almost every industry, very demand are studies of classification methods that, prior to their introduction into decision support systems. Compare analysis of the secant planes method, method of the potentials and potential method in the field of receptors are shown in the paper. At first, in introduction, authors shew needs of autonomic systems of adaptive perception on visible diapason of specter. As particularly aim, these methods are compared by criteria of speed, accuracy and amount of storage used after training. As general idea we are looking for we are looking for methodic of the best combination of method for different condition on observe field of visual spectral diapason. Theories of the every method are presented, and then tables of compare analysis of results are shown. Step-by-step comparative experiments are described in detail. Changes at each step are shown in detail in the tables of the corresponding signs. Moreover, at the end of the paper, comparative characteristics of each method with the same learning time in same type of experiments for each method are presented. As a result, in the first group of tables , we see a difference in the recognition time and the amount of memory required for correct operation. Those are truth tables for two points, three points, two points and two planes, three points and two planes, three points and three planes, three points and seven planes. The conclusion gives a thorough explanation of where to use the best method. The needs of the system for computing resources in the application of each mode are presented and corresponding dependencies are derived. Next, If you train several times on the same object (ie, train several times), you can expect that the errors in the breakdown of the receptor space will be different. In this case, you can improve the performance of the algorithm by parallelizing its process into several threads. Using this method simultaneously and independently of each other on the same image is multi-threaded learning on multiple computer kernels. When recognizing new objects, they will refer to some image, not necessarily the same. The final decision is made by "vote" - the object refers to the image to which it was attributed to a greater number of parallel streams.