Inland lakes or reservoirs are the main surface water resources, related to people’s survival and social development. On-site water level data are critical for water resource management, while satellite laser and remote sensing data are a significant tool for long-term change monitoring. In this study, we took the Miyun Reservoir, the most important source of drinking water in Beijing, as the research area. We used Landsat images and the Ice, Cloud, and Land Elevation Satellite (ICESat-2) for small-scale water level change monitoring and inversion across unrecorded periods. First, we proposed a new two-step denoising method based on local statistics to denoise the ATL03 single photon data from 2018 to 2020, and extracted the effective elevation information of Miyun Reservoir. Second, based on Landsat images from 1984 to 2020, the MNDWI index was used to extract the water area and generated a percentage map of water occurrence. Finally, the water and land boundary elevation were selected from the effective photon elevation information according to the boundary position under each percentage of the water generation percentage graph, the area, and the water level elevation model (A–E model) was established, and the water level values of unrecorded periods were inverted. The denoising results showed that the denoising accuracy of all data is higher than 95%. The area of Miyun Reservoir has changed greatly in the past 37 years, especially in the northeast corner and northern region. The inverted water level elevation had a strong correlation with the measured value, the correlation coefficient reached 0.97, the root mean square error (RMSE) was 0.553 m, and the estimated water level value was highly consistent with the recorded value and fluctuated dramatically from 1984 to 2020. This method complements the traditional water level mapping method, provides a new idea for the monitoring of water level changes in lakes or reservoirs without field records, and also contributes to the timely and effective monitoring and management of global lake and reservoir water level and time changes.

Stripe noise is considered one of the largest issues in space-borne remote sensing. The features of stripe noise in high-resolution remote sensing images are varied in different spatiotemporal conditions, leading to limited detection capability. In this study, we proposed a new detection algorithm (LSND: a linear stripe noise detection algorithm) considering stripe noise as a typical linear target. A large-scale stripe noise dataset for remote sensing images was created through linear transformations, and the target recognition of stripe noise was performed using deep convolutional neural networks. The experimental results showed that for sub-meter high-resolution remote sensing images such as GF-2 (GaoFen-2), our model achieved a precision of 98.7%, recall of 93.8%, F1-score of 96.1%, AP of 92.1%, and FPS of 35.71 for high resolution remote sensing images. Furthermore, our model exceeded ~40% on the accuracy and ~20% on the speed of the general models. Stripe noise detection would be helpful to detect the qualities of space-borne remote sensing and improve the quality of the images.

Oblique stripe noise widely appears in remote sensing images after image correction, exhibiting arbitrary tilt angles and parallel distribution. Due to its arbitrary randomness in tilt angles and lengths, oblique stripe noise increases the difficulty of detection compared to vertical or horizontal stripe noise. For the first time, we propose a group of oblique stripe noise synthesis and detection algorithms combining imaging mechanisms and deep learning. To get controllable synthetic oblique stripe noise data for training detection model, two sample augmentation methods are presented by the image correction’s imaging mechanisms with new linear transformation and the generative adversarial network algorithm with Cycle-GAN, respectively. A large-scale simulated stripe noise dataset (SOSD, simulated oblique stripe noise dataset) is simulated using these two methods. A new deep learning detection algorithm (RDOS, Robust detection of oblique stripe Noise) is presented considering the presence of oblique stripe noise. RDOS is trained using both SOSD and a real stripe noise dataset, and it obtains the optimal detection model for testing. The experimental results show that the accuracy reaches 82.93%, the recall rate reaches 85.17%, the F1 score reaches 84.04%, the average precision (AP) reaches 82.34%, and the frames per second (FPS) reaches 33.33. Compared with the general line detection models, our model exceeds ~300% in accuracy and ~60% in speed. In the future, the proposed algorithms have great potential for application in various areas such as quality evaluation, image preprocessing, and engineering problems related to multi-angle linear object augmentation and detection.

In order to improve the image registration accuracy of heterogeneous remote sensing images with large radiation differences, an improved image registration algorithm for thermal infrared and panchromatic images is proposed. This method uses the phase consistency of intensity and direction to construct a geometric structure feature descriptor called the Histogram of Oriented Phase Congruency (HOPC) for image registration. It employs the Curvature Scale Space (CSS) corner detection to concentrate and extract feature points in contour areas, and defines a similarity metric (called HOPCn) based on the Euclidean distance between the descriptors for high-precision heterogeneous images registration. Five sets of panchromatic and thermal infrared images were selected for verification. The results show that the improved algorithm can match the homonymy points around the image contour, and the Correct Matching Rate (CMR) is improved by 2.7% and the Root Mean Square Error (RMSE) is improved by 3.7% compared to the original algorithm. This proves that the proposed method has good robustness under conditions of large radiation differences.