Automatic detection of vertex defects in low-light-level image intensifiers based on spectral maximum points histogram statistics and analysis of multifilament boundary distribution patterns.

Abstract

Vertex defect is a defect introduced during the physicochemical treatment phase of the microchannel plate in low-light-level (LLL) image intensifiers. It will affect the imaging quality of the product when the contrast exceeds a certain range, so it should be detected in time before delivery. Traditional detection techniques for this defect are divided into visual inspection and objective detection. High degree of uncertainty is the main drawback of the subjective method, while the existing objective detection technology has limitations in terms of detection coverage. To address the above problems, an automatic detection method for vertex defects based on spectral maximum points histogram statistics and analysis of multifilament boundary distribution patterns is proposed. First, a spatial bandpass filter is adopted to enhance visible defects in the effective area. After performing image binarization and connected component analysis, the regions corresponding to visible defects are extracted, and the processed image is generated. Then, a sliding window of equal multifilament size is utilized to scan the effective area of the processed image globally. By performing histogram statistics on the local maximum point coordinates of the spectrogram, defective image blocks can be identified using the maximum weight bin of the histogram. Subsequently, defective image blocks are restored to the processed image to obtain dense areas of vertex defects. Finally, the distribution pattern of the multifilament boundary is mathematically analyzed and defined as two constraints that can determine the effective line segment composed of vertex defects. To substantiate the performance of this method, multiple products are put into the experiment and relevant defect detection techniques are used for comparison. The experimental results indicate that our method has better detection performance than relevant objective detection methods, and can generate results consistent with the visual inspection method, while maintaining applicability to changes in image brightness, clarity, contrast, background noise, and uniformity within a certain range.