Several Technique Research For The Redundancy Detection On Complicated Component Image

In the industrial society, the integrality of devices is the important guarantee of the working efficiency of machines. Thus, it is essential to develop methods to detect the redundancy in devices. Currently, most of the redundancy detection for device depends on human check or human-machine check, which make the result has much subjectivity. Besides, in the human check, the technician is prone to be tired, which might lead to the drop of the working efficiency. Aiming at this problem, we advance a new redundancy detection method, which based on the pattern recognition and vision theory.After the careful analysis for the image of device, we find the different features of redundancy and the background of image. Redundancy usually have small size, different shapes and flexible positions. on the contrary, the background of image usually has several different regions, and the shape, size, position of regions are relatively steady. In view of these features, we propose to use the recognition and registration of the marked region in the background to complete the registration between reference image and real-time image.The primary work of this paper is to fix the position of redundancy in the device. For attaining this purpose, we did work in several aspects as below: (1) image segmentation; (2) the feature extraction in the region; (3) the selection of the marked features; (4) the recognition of the marked region; (5) the registration of the marked region; (6) the detection of redundancy. Most of the target recognition methods are designed as the feature of target, in contrast, this paper make the redundancy recognition from the recognition of the region in background. After the pre-processing and segmentation of the images, according to the extracted features from regions, we find the most marked regions and then, make the registration between the marked regions. At last, detect the redundancy in devices. This paper made test in four kinds of devices, which have 300 images. Experiments proved that the feasibility of this method and the detection probability can reach above 0.85.