Research On ML-FRFT Algorithm And The Application Of Instrument Image Recognition

The automatic recognition of instruments is the core technology to achieve the remote monitoring function of instruments. When we recognize the instrument character under the condition of uneven illumination, the traditional template matching method is not suitable to meet the needs of practical application. So this study proposes a whole-information feature extraction and recognition method based on line-segment detection, in order to improve the accuracy of the instrument image under the condition of non-uniform background light of an image.To extract the line effectively, this paper uses the multilayer fractional Fourier transform(ML-FRFT) for the detection of straight line. Compared with the Hough transform and Radon transform, the results shows that ML-FRFT line detection algorithm can accurately detect the different width of lines in an image, as well as the image of as many as dozens of the intersection of straight line. And its precision is better than the Hough Transform and Radon Transform. What’s more, ML-FRFT is robust and not sensitive to noise. The most important is this algorithm can make a line detection of gray image and even colorful image directly with no need to convert image into binary image for edge information.The precision of ML-FRFT about line detection image is high, but when coming to extract the feature information, getting line segment is necessary, which requires the straight line detection on the basis of further detecting line endpoint. This paper studies the line segment detection algorithm based on ML-FRFT, adding a window to the line direction that is under detection to get the line’s endpoint by checking the edge of sinogram-butterfly-wave. It is proved that this method is able to detect line segments of the binary image effectively, but it can’t detect line endpoint of gray image. At the same time, the classical Hough Transform and LSD algorithm is studied by comparing experiments, which proves that LSD algorithm can accurately detect gray image edge line, but the line will be cut off. However, the Hough Transform has small deviation and the segment which detected may be shorter than the real one because of the dependence of edge detection. As the line segment detection algorithm based on ML-FRFT is not suitable for grayscale image, this paper proposes a ML-FRFT based on threshold limit line segment detection algorithm, successfully detect line segment in gray image.In conclusion, the Hough Transform algorithm is used after testing about image of instruments. Firstly, with the pre-processing of image by filtering, detect line character to make the line tilted. At the same time, match detected line segments to extract all the pixels feature. Finally, use the neural network algorithm for character recognition. It can be proved that this system can effectively improve the accuracy of automatic recognition system of instruments in uneven illumination.