When sapphire crystal is prepared with Kyropoulos method, the necking-down growth process is a key stage. Sapphire growth defect is a big problem in this stage. However, diagnosing growth defects is subject to the interference of workers subjectivity and accuracy always goes down. To address the problem, a novel defect diagnosis method is proposed for necking-down growth process in this paper. Industrial CCD sensors replace eyes of skilled workers to observe in this method. A new Defect-Diagnosing Siamese network (DDSN) is used in this method. We use Siamese architecture to learn similarity through pairs of images. We use the deep separable convolution (DSC) into the DDSN to optimize running speed and model size. In experiment, dataset is acquired by industrial CCD sensors in the necking-down growth process. The accuracy of defect diagnosis can reach up to 94.5%. The method significantly improves the traditional way.

Multi-modal remote sensing images often exhibit complex and nonlinear radiation differences which significantly hinder the performance of traditional feature-based image registration methods such as Scale-Invariant Feature Transform (SIFT). In contrast, structural features—such as edges and contours—remain relatively consistent across modalities. To address this challenge, we propose a novel multi-modal image registration method, Cof-SIFT, which integrates a co-occurrence filter with SIFT. By replacing the traditional Gaussian filter with a co-occurrence filter, Cof-SIFT effectively suppresses texture variations while preserving structural information, thereby enhancing robustness to cross-modal differences. To further improve image registration accuracy, we introduce an extended approach, Cof-SIFT_HOG, which extracts Histogram of Oriented Gradients (HOG) features from the image gradient magnitude map of corresponding points and refines their positions based on HOG similarity. This refinement yields more precise alignment between the reference and image to be registered. We evaluated Cof-SIFT and Cof-SIFT_HOG on a diverse set of multi-modal remote sensing image pairs. The experimental results demonstrate that both methods outperform existing approaches, including SIFT, COFSM, SAR-SIFT, PSO-SIFT, and OS-SIFT, in terms of robustness and registration accuracy. Notably, Cof-SIFT_HOG achieves the highest overall performance, confirming the effectiveness of the proposed structural-preserving and corresponding point location refinement strategies in cross-modal registration tasks.

Land surface temperature (LST) is a fundamental variable of environmental monitoring and surface equilibrium. Although the HJ-1B infrared scanner (IRS) has accumulated many observations, further application of HJ-1B/IRS is limited by the lack of LST products. This study refined the ERA5 atmospheric profile database, instead of the widely used traditional TIGR atmospheric profile database, and simulated the coefficients of the generalized single-channel (GSCs) algorithms to improve LST retrieval. GSCs can be divided into the GSC_w and GSC_wT algorithms, depending on whether the input is atmospheric water vapor content (WVC) or in situ near-surface air temperature and WVC. Land surface emissivity (LSE) was obtained from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Emissivity Dataset (GED) and vegetation/snow cover products. Then, the retrieved LSTs were evaluated using the LSTs from the RTE algorithm, TIGR_w/TIGR_wT profiles, and in situ near-surface air temperature from the HiWATER experiment in China from 2012 to 2014. The bias (root mean square error (RMSE)) values are displayed as ERA5_wT < RTE < ERA5_w < TIGR_wT < TIGR_w. The accuracy of ERA5_wT, with a bias (RMSE) of 0.02 K (2.30 K), is higher than that of RTE, with a bias (RMSE) of 0.74 K (2.47 K). The accuracy of RTE is preferable to that of ERA5_w, with a bias (RMSE) of 0.89 K (2.48 K), followed by TIGR_wT, with a bias (RMSE) of −1.18 K (2.50 K), and then, TIGR_w, with a bias (RMSE) of 1.60 K (2.77 K). In summary, the accuracy of LST obtained by GSC from the refined ERA5 atmospheric profiles is higher than that obtained from the TIGR profiles. The accuracy of LST obtained by GSC_wT is greater than that obtained by GSC_w. The accuracy of LST obtained using in situ near-surface air temperature is higher than that obtained using ERA5 air temperature. The accuracy of LSE_ASTER is slightly better than that of LSE_MOD21. The aforementioned conclusions can provide scientific support to generate HJ-1B/IRS LST products.