Signal modulation classification (SMC) has attracted extensive attention for its wide application in the military and civil fields. The current direction of combining deep learning technology with wireless communication technology is developing hotly. Deep learning models are riding high in the field of SMC with their highly abstract feature extraction capability. However, most deep learning models are decision-agnostic, limiting their application to critical areas. This paper proposes combining traditional feature-based methods to set appropriate manual features as interpretable representations for different modulation classification tasks. The fitted decision tree model is used as the basis for the decision of the original model on the instance to be interpreted, and the trustworthiness of the original deep learning model is verified by comparing the decision tree model with the prior knowledge of the signal feature-based modulation classification algorithm. We apply the interpretable explanation method under the current leading deep learning model in the field of modulation classification. The interpretation results show that the decision basis of the model under a high signal-to-noise ratio(SNR) is consistent with the expert knowledge in the traditional SMC method. The experiments show that our method is stable and can guarantee local fidelity. The decision tree as an interpretation model is intuitive and consistent with human reasoning intuition.

In recent years, weakly supervised semantic segmentation (WSSS) has garnered significant attention in remote sensing image analysis due to its low annotation cost. To address the issues of inaccurate and incomplete seed areas and unreliable pseudo masks in WSSS, we propose a novel WSSS method for remote sensing images based on the Siamese Affinity Network (SAN) and the Segment Anything Model (SAM). First, we design a seed enhancement module for semantic affinity, which strengthens contextual relevance in the feature map by enforcing a unified constraint principle of cross-pixel similarity, thereby capturing semantically similar regions within the image. Second, leveraging the prior notion of cross-view consistency, we employ a Siamese network to regularize the consistency of CAMs from different affine-transformed images, providing additional supervision for weakly supervised learning. Finally, we utilize the SAM segmentation model to generate semantic superpixels, expanding the original CAM seeds to more completely and accurately extract target edges, thereby improving the quality of segmentation pseudo masks. Experimental results on the large-scale remote sensing datasets DRLSD and ISPRS Vaihingen demonstrate that our method achieves segmentation performance close to that of fully supervised semantic segmentation (FSSS) methods on both datasets. Ablation studies further verify the positive optimization effect of each module on segmentation pseudo labels. Our approach exhibits superior localization accuracy and precise visualization effects across different backbone networks, achieving state-of-the-art localization performance.

Absolute visual localization based on onboard cameras and georeferenced data is the dominant solution for unmanned aerial vehicles (UAVs) when the global navigation satellite system is not available. However, current image-based matching methods are easily affected by significant feature differences between UAV images and georeferenced images, and map-based matching methods are affected by image rotation and resolution differences. To address the above-mentioned challenges, this article proposes a rotation and scale invariance map matching method (RSIM) to realize UAV visual localization in urban scenarios. Specifically, RSIM first extracts building contour information from UAV images. Then, based on the shape and spatial relationship features of the buildings, a scene matching method is designed to match the UAV image and the vector e-map. Finally, the centers of the matched building individuals are used as control points for UAV position solving. Extensive experiments conducted on the LoFTR and DKM datasets demonstrate that the proposed RSIM algorithm achieves superior localization performance compared to existing deep learning-based image matching methods. Moreover, our RSIM effectively enables UAV localization in urban scenarios even when the orientation and resolution of UAV images are unknown, achieving performance comparable to the SSRM map matching method, which relies on prior knowledge of these parameters.