Research On The Key Technologies In Insulator Defect Detection Based On Image

Insulator is a necessary device of transmission line and it is essential to maintain its safe and stable operation. Analyzing insulator using image processing technology and realizing automatic defect detection, which can provide accurate decision support, have practical significance for transmission line automatic inspection. Taking visible and infrared insulator images as data source, deep studies on relevant key technologies, including denoising, segmentation, recognition and defect detection, were conducted. The main work is as follows.In order to achieve better denoising effect of high density impulse noise image, an adaptive algorithm based on modified peer groups was proposed. The size of impulse noise point’s neighborhood window grew adaptively according to the number of non-impulse noise point in it to obtain the peer groups and combination of median filter and mean filter were used to remove impulse noise based on the number of the peer groups. Experiments on the standard test images verified that the proposed method can obtain better denoising effect than the other methods.A denoising algorithm based on modified fuzzy peer groups is proposed for mixed impulse and Gaussian noise. The proposed algorithm for removal of high density impulse noise was used to remove the impulse noise in mixed noise. Variance estimation method for Gaussian noise based on nonsubsampled contourlet transform and principal component analysis is proposed, whose result was combined with fuzzy peer groups to remove Gaussian noise. The effect of proposed algorithm is validated by experiments on both standard test images and actual visible and infrared images.To obtain better segmentation effect, a modified unit-linking pulse-coupled neural networks (MUL-PCNN) image segmentation algorithm is proposed. The linking strength coefficient is determined by computing the variances of each neuron center and the mean square error(MSE) is used to select the optimal image to realize better effect of segmentation. On standard test image peppers and visible and infrared insulator images from the scene, experiments are carried out and the results indicated the proposed algorithm can get better segmentation effect for insulator images and have better robustness.A recognition method of binarization morphology and affine scale-invariant feature transform operator was given to solve the multi-scale and multi-angle problem in recognition process of insulator. MUL-PCNN algorithm is used to segment insulator images to gain binary image and morphology algorithm is used to divide the binary image into several regional blocks to be detected, ASIFT matching was done between each block and the template of insulator string and the insulator string can be recognized according to the number of matching point. Experiments on several on-site insulator images indicated that the proposed method can recognize the insulator string accurately.String breakage defect detection based on sparse representation was proposed. Insulator’s spindle was determined by progressive corrosion thought and then be rotated to vertical mode, which was projected to axis coordination for statistic to determine the height and width of single disc and then to extract information of it. String breakage defect detection was realized by classifying the extracted single discs. Higher precision of the method was confirmed by the experiments on the aerial insulator images.