The geological environment remote sensing interpretation task imposes higher demands on deep learning models' computational efficiency and deployment performance. As an efficient lightweight strategy, channel pruning has been widely applied in computer vision. However, when applied to geological remote sensing tasks, existing pruning methods struggle to balance interpretation accuracy and computational cost effectively, mainly due to the complexity of geological environment data and the constraints of resource-limited deployment. Moreover, although reinforcement learning-based pruning strategies have been proposed to overcome these limitations, they often suffer from limited exploration capability and unstable training performance. To address these issues, we propose a dynamic pruning method (SACRL) incorporating the maximum entropy mechanism, which enhances the explorability and stability of the pruning strategy by introducing an entropy regularity term in the reinforcement learning objective function. Experiments demonstrate that, compared to conventional pruning methods such as AMC and Taylor, SACRL demonstrates better performance in terms of overall classification accuracy (OA), mean Intersection over Union (mIoU), and mean Average Precision (mAP). Furthermore, deploying the pruned model on resourcelimited devices, such as the Jetson AGX Orin, further validates its adaptability in real-world applications and highlights its potential in remote sensing interpretation of geological environments.

Three-dimensional (3D) reconstruction based on aerial images of unmanned aerial vehicles (UAVs) is important for surveying and monitoring natural resources, supporting land cover analysis and geological hazard mapping. However, the performance of a fixed UAV docking station and manual deployments in large areas is limited by slow data acquisition and transmission. To accelerate data acquisition, emerging mobile UAV docking stations can collaboratively capture images in wide areas. Moreover, by leveraging edge computing power on UAVs and docking stations, the computations of 3D reconstruction can be performed locally, eliminating the need for data transmission. This paper builds a mathematical model for the entire process. The model can be decomposed into three associated problems: area partitioning, computation offloading, and mobile docking station path planning. The optimization objective is to minimize the execution time under the constraint of UAV battery power. A suboptimal solution is first derived using an enumeration-genetic algorithm and then fine-tuned using deep reinforcement learning. The experimental results validate the feasibility of using mobile docking stations for 3D reconstruction. In addition, the numerical results indicate that our proposed solution reduces execution time compared to the benchmark solution.

Preemptive scheduling reallocates resources to highpriority pods by terminating low-priority pods when idle resources are insufficient in a Kubernetes cluster. The default Kubernetes preemption strategy suffers from poor resource utilization because it neglects scoring metrics, such as network bandwidth and image locality. In this paper, an efficient preemptive resource scheduling method is proposed to address this issue with a well-designed two-layer-based pod prioritization mechanism considering both the pod's restart policy and runtime resource usage, and a node scoring model taking the network bandwidth into consideration. Experimental results show that the proposed method significantly improves resource utilization, reduces scheduling latency, and increases the success rate of highpriority pod deployments compared to the default Kubernetes strategy.