In the process of microchannel plate (MCP) making and physicochemical treatment of a low-light-level (LLL) image intensifier, multifilament fixed pattern noise, also known as structural defects, is one of the most common defects in the anode surface. The appearance of this defect will seriously affect the imaging quality of an image intensifier, so it should be found in time before delivery. The traditional evaluation method of this defect relies on subjective judgment, and the disadvantage is that the division of the dense defect area and the measurement of defect gray difference (GD) are not standardized. To address this problem, an automatic evaluation method of vertex structural defects of an LLL image intensifier based on proposed individual image processing strategies is presented, which provides a digital evaluation scheme for such defects. This method is composed of two parts: quasi-circular defect detection and defect GD calculation. The first part is composed of coarse detection and fine detection. Coarse detection is to scan the anode surface and take the two ends of a pair of adjacent line segments with a large gradient sum and opposite gray change direction as the defect boundaries; fine detection is to establish the image patch from defect boundaries, extract the edge segment from the image patch, and judge whether it conforms to the shape of a circle. In order to substantiate the performance of the quasi-circular defect detection strategy, two relevant techniques are used as comparison. One is based on a Gaussian filter, and the other is based on a fixed-size window template. The comparison results show that our method, to the best of our knowledge, has the best detection performance for vertex structural defects. The second part consists of region of interest (ROI) cropping, secondary defect detection, shortest distance sequence establishment, effective distance extraction, triplet set construction, and triplet GD calculation. First, the location histogram of defects is established to cut ROI; then, the secondary defect detection is performed to extract more vertex structural defects from ROI; after that, the shortest distance sequence of defects is constructed, and the effective distances are extracted by using the structural features of multifilament. Finally, the triplet set is generated according to the effective distance, and the triplet GD is calculated based on the gray information near the triplet baseline. The GD of vertex structural defects corresponds to the maximum GD of triplets. So as to verify the effectiveness of vertex defect GD calculation strategy, several image tubes with different degrees of such defects are used for experiments, and the subjective evaluation method is used as comparison. The experimental results substantiate that this method is superior to the subjective method in locating ROI accurately and calculating defect GD quantitatively. In general, the automatic evaluation method can be regarded as an effective evaluation scheme for vertex structural defects of an LLL image intensifier.

No abstract

No abstract

Siamese networks have attracted wide attention in visual tracking due to their competitive accuracy and speed. However, the existing Siamese trackers usually leverage a fixed linear aggregation of feature maps, which does not effectively fuse the different layers of features with attention. Besides, most of Siamese trackers calculate the similarity between the template and the search region through a cross‐correlation operation between the features of the last blocks from the two branches, which might introduce the redundant noise information. In order to solve these problems, this study proposes a novel Siamese visual tracking method via cross‐layer calibration fusion, termed SiamCCF. An attention‐based feature fusion module is employed using local attention and non‐local attention to fuse the features from the deep and shallow layers, so as to capture both local details and high‐level semantic information. Moreover, a cross‐layer calibration module can use the fused features to calibrate the features of the last network blocks and build the cross‐layer long‐range spatial and inter‐channel dependencies around each spatial location. Extensive experiments demonstrate that the proposed method has achieved competitive tracking performance compared with state‐of‐the‐art trackers on challenging benchmarks, including OTB100, OTB2013, UAV123, UAV20L, and LaSOT.