Research On Surface Defect Inspection Of Strip

Strip surface detection technology makes a big different of the billet production process. on the one hand, the production chain can be adjusted to improve billet’s quality according to the detected defects, on the other hand,billet can be classified according to the test results, positioning billet quality accurately help reduce trade disputes. The high temperature, mechanical vibration, uneven illumination in the production line on site put forward high requirements for the strip surface detection system. This paper introduced a design of the entire system which make up with a high-speed line scan cameras as the acquisition equipment, FPGA as transmission equipment and industrial PC as parallel processing equipment. The system is able to acquire high-resolution images instantaneously and steadily. In the process of image detection and identification this paper has done the following studies.This paper proposed a negative power transform homomorphic filter to remove the spot on the live images which come into being by the uneven illumination. Compared with the exponential and logarithmic transform homomorphic filter, negative power transform homomorphic filter has better spatial dynamic range compression and image contrast enhancement effect. At the same time, the filter has a adjustable parameters, which can satisfy different degrees of homomorphic filtering requirements.Inspired by Canny operator, this paper proposed a edge detection algorithm named dual-threshold bandpass Sobel edge detection operator. Setting reasonable dual threshold of this operator can detect the defect edge with no omissions. The operator combines morphological filtering to get rid of a lot of noise. Compared to Canny operator, the operator is able to extract the defect edge within any given gradient range.In the image recognition algorithm, a feature vector was a neural network’s input. This paper compared the effect of the BP neural network, RBF neural network recognition and inputs standardized RBF neural network algorithm, and did the assessment of the three neural networks.The effect of the detection algorithm was determined by the parameters of the dual-threshold. In order to make the parameters in the detection operator more accurate, feedback correction algorithm was proposed. This algorithm is used to make parameters in detection link has dynamic adjustment ability.