Strain Gauge Image Feature Extraction And Identification

Pattern recognition and machine vision technology have been applied to more and more automated installation systems. But due to the low efficiency, low accuracy and poor consistency of artificial installation strain gauge, the automatically installed strain gauge system design and research was put on Schedule. Strain gauges in drawing process prone to deflection, drift, thus it affects the accuracy of the system. So in order to preprocess strain gauge image effectively, feature extraction and pattern recognition will affect system performance. This thesis concentrates on how to extract the feature so strain gauges by effective algorithm and using the classification of rapid identification of the strain gauges.In this thesis, two methods have been discussed below, such as how to use the fast and efficient algorithm to extract the features of the strain gauge and how to recognize the strain gauge by using the fast classification are discussed.In the process of extracting the feature of the strain gauge, this thesis introduces the research work on the conventional edge detection and corner detection which has been applied in the strain gauge image. Due to the practical application needs, such as reducing the impact of noise and restraining the background noise, the structure element probe is introduced to reduce the impact on the complexity of classification. The edge detection algorithm based on structure element is also introduced. It detects the strain gauge edge by using the structure elements in four directions; the priority of this algorithm has low computation load of and effective reduction of the noise and the edge of the background. The advance SUSAN detection algorithm is aim at confronting the complexity of conventional SUSAN detection ones. Because the change of the foreground and background in stain gauge is not obvious, the result of the corner detection is not good.The feature of the strain gauge should be rotation invariant in auto installation system. The moment invariant algorithm which includes Hu moment, Zernike moment and wavelet moment has been discussed. The Hu moment and Zernike moment is counted in the whole space of the image, so they can be easily interrupted by the noise. The whole moment and local feature moment of the strain gauge can be got in the same time by using the multi-scale analysis of wavelet moment. The result shows that it can restrain the noise, and it is fit for the strain gauge auto installation system.The samples of the stain gauge are limited in practical stain gauge auto installation system; therefore the recognition may not be correct by using image recognition, like template matching. BP neural net may got the local minimum and it's converge speed is slow and the process of the parameter initialization in the wavelet neural net is complex, so they cannot satisfy the time and accuracy requirements of the stain gauge auto installation system. SVM can deal with these problems, such as high dimensions, linearity and local minimum. Because of these priorities and fast computation, the recognition combining SVM and the wavelet moment and corner is introduced in this thesis. This method is easy and efficient and is able to meet the requirements of system time and accuracy.