Research On Enhancement And Recognition Method Of Low Illumination Train Fault Image Based On Sparse Coding Based Spatial Pyramid Matching

Trouble of Moving Freight Car Detection System(TFDS)is a fault image detection system based on machine vision developed by china.Because of various types of TFDS faults and collection environment of all weather,the collected images are prone to problems such as complex background,low illumination and high noise.In view of the above problems,a low illumination train fault image enhancement and recognition method is proposed based on sparse coding based spatial pyramid matching and genetic algorithm optimized support vector machine(GA-SVM).By enhancing the low illumination TFDS image and extracting multi-scale coding features,the automatic recognition of typical train faults under low illumination is realized.From the human visual system(HVS)model,an image enhancement algorithm is proposed based on accurate illumination image estimation.Firstly,HVS model was used to establish the decomposition relation of the image,and the mapping relationship is constructed by finding the maximum value in the three color channels of R,G and B of images.Then,accurately estimate the illumination image by the mapping relationship of illumination structure combined with the first derivative filtering and augmented lagrangian method,and the enhanced image is obtained according to the Retinex theory.Furthermore,the BM3 D method is used to denoise the amplified noise in the reflected image.Finally,the final result is obtained by fusing the denoised image and the enhanced image.The experimental results show that the proposed algorithm is faster.In the same speed level comparison,the proposed algorithm performs better than the traditional method in four image evaluating index such as peak signal to noise ratio(PSNR),structural similarity index measure(SSIM),lightness-order-error(LOE),gradient magnitude similarity deviation(GMSD).The enhanced image is more prominent in subjective visual color saturation and contrast,and the edge contour details are clear,which meets the actual needs of image enhancement.In view of the variety of TFDS image fault types,a general fault automatic recognition algorithm is proposed based on sparse coding based spatial pyramid matching and GA-SVM.Firstly,the image was divided into patch areas in different scale spaces,and the scale-invariant feature transform(SIFT)of the patch areas were extracted.Sparse coding is performed by iteratively learning dictionaries using the SIFT features of randomly extracted samples.Secondly,the principal component analysis was used to define the contribution of the encoded features to the fault recognition accuracy,and reduce the dimensionality of the coding features.Then,the SVM classifier is trained by the feature of reduced dimension after coding and genetic algorithm.Finally,the trained classifier was used to detect the bogie block key,dust collector and fastening bolt faults.The experimental results show that the algorithm can adapt to the recognition for three different kinds of faults.The fault recognition rate was 97.25%,99.00% and 97.50% respectively.It can meet the actual requirements of detection to vehicle's faults.According to the actual detection environment of TFDS,the fault recognition of bogie block Key(BBK),dust collector(DC)and fastening bolt(FB)under low illumination scene is realized by combined with the image enhancement algorithm based on accurate illumination image estimation and the fault automatic recognition algorithm based on sparse coding based spatial pyramid matching and GA-SVM.The proposed algorithm has high recognition accuracy and good robustness.It can be effectively applied to TFDS fault detection in complex environments with low illumination.