Unmanned aerial vehicles (UAVs) have been widely used in aerial photography and target detection tasks due to their flexibility and unique perspective. However, small targets often suffer from insufficient resolution, uneven scale distribution, and complex background clutter, which are constrained by imaging conditions such as high-altitude imaging, long-distance capture, and wide field of view. These factors weaken the feature representation and generalization ability of the model, becoming the key bottleneck that restricts the improvement of small target detection accuracy in UAV scenarios. To address the above issues, this paper proposes a small target detection algorithm for UAV perspective, namely MTD-YOLO. First, a Parallel Multi-Scale Receptive Field Unit (PMSRFU) is designed. This unit effectively enhances the receptive field range of feature extraction and the fusion ability of multi-scale contextual information by introducing parallel branches with different-sized convolutional kernels. Second, we embed PMSRFU into a C2f block to form C2f-PMSRFU, which reuses shallow details and fuses multi-scale features to clarify edges and textures in small targets, yielding stronger fine-grained representations. Finally, an efficient detection head with task decoupling, dynamic alignment, and adaptive scale adjustment capabilities, namely SDIDA-Head, is proposed, which significantly improves the model’s small target detection accuracy. Extensive experiments on the VisDrone2019 and HazyDet datasets demonstrate that MTD-YOLO achieves a 7.6% and 6.6% increase in mAP@0.5 compared to the baseline YOLOv8n, respectively. Meanwhile, the Precision is improved by 6.0% and 1.1%, and the Recall is enhanced by 7.5% and 6.9%, respectively. These results fully validate the effectiveness and superiority of the proposed method in UAV small target detection tasks.

Ground-based cloud image features high-spatiotemporal resolution, presenting detailed local cloud structures and valuable weather information, which are crucial for meteorological forecasting. However, the inherent fuzziness and dynamism of ground-based clouds have hindered the development of effective prediction algorithms, resulting in low accuracy. This paper presents CloudPredRNN++, a novel method for predicting ground-based cloud dynamics, leveraging a deep spatiotemporal sequence prediction network enhanced with a self-attention mechanism. Initially, a Cascaded Causal LSTM (CCLSTM) with a dual-memory group decoupling structure is designed to enhance the representation of short-term cloud changes. Next, self-attention memory units are incorporated to capture the long-term dependencies and emphasize the non-stationary characteristics of cloud movements. These components are integrated into cloud dynamic feature mining units, which concurrently extract spatiotemporal features to strengthen unified spatiotemporal modeling. Finally, by embedding gradient highway units and adding skip connection, CloudPredRNN++ is constructed into a hierarchical recursive structure, mitigating the gradient vanishing and enhancing the uniform modeling of temporal–spatial features. Experiments on the sequence ground-based cloud dataset demonstrate that CloudPredRNN++ can predict the future cloud state more accurately and quickly. Compared with other spatiotemporal sequence prediction models, CloudPredRNN++ shows significant improvements in evaluation metrics, improving the accuracy of cloud dynamics forecasting and alleviating long-term dependency decay, thus confirming the effectiveness in ground-based cloud prediction tasks.

Cloud detection in remote sensing imagery is a research hotspot in the field of image processing, and accurately detecting and segmenting cloud regions is crucial for improving the utilization efficiency of remote sensing data. However, standard convolutional neural networks face limitations in modeling the complex spatial structures of clouds. To address these challenges, this paper proposes a cloud detection method based on DDS-DeeplabV3+. First, a lightweight design of the Xception network is adopted to control model complexity, and part of its standard convolutional layers are replaced with Deformable Convolutional Networks (DCN), which enhances the capability of the model to capture geometric features of irregular cloud formations. Second, a Dual-Branch Collaborative Mechanism (DCM) that integrates global context modeling with local detail perception is designed to reconstruct the Atrous Spatial Pyramid Pooling (ASPP) module, thereby improving performance in handling complex scenes and fine boundary delineation. Finally, the SimAM (Simple, Parameter-Free Attention Module) is incorporated into the decoder module, enhancing thin cloud detection capability. Experimental results on the Landsat-8 and GF-1 datasets show that the proposed model achieves Mean Intersection over Union (MIoU) values of 92.61% and 94.04%, respectively, outperforming other comparative methods and demonstrating its superior performance in cloud detection tasks.