Recently, the use of Unmanned Aerial Vehicle (UAV) imagery for object detection in forest fire detection has gained significant attention and has shown remarkable performance. However, most existing object detection models have neglected the exploration of relationships between positive sample features, which is crucial for learning more representative and color-robust features. Additionally, small objects in UAV images poses challenges in capturing sufficient object information and hinders accurate object detection. To address these issues, we propose FCLGYOLO that aims to constrain positive sample features and enrich the object information in the feature maps. Specifically, a Feature Invariance and Covariance Constraint (FICC) structure proposed to maintain feature invariance among positive samples and remove internal correlations. Furthermore, a Local Guided Global Module (LGGM) proposed to enrich object positioning and semantic information in the feature map, which leverages local features that focus on spatial information to facilitate the learning of global features that focus on frequency information. It is interesting to show that FCLGYOLO performs well even in the presence of heavy smoke or tree occlusions. Compared with multiple state-of-the-art object detection models on a forest fire dataset, experimental results demonstrate the superiority of FCLGYOLO. The code is available at: https://www.github.com/zhangshao249/FCLGYOLO .

Recently, self-supervised pretraining algorithms have been extensively investigated in the field of remote sensing, yielding significant performance improvements in change detection tasks. However, many existing algorithms inadequately exploit multi-level features, which impedes their ability to precisely extract salient change characteristics and accurately determine the structural integrity of changed targets. Moreover, most mainstream approaches tend to neglect the mining of shallow, frequency-domain details, thereby limiting the model's capacity to discern object boundaries. To address these challenges, we propose a novel multi-dimensional feature alignment self-supervised pretraining framework, termed MDFANet. Specifically, we introduce a Hierarchical Decoupling and Multi-dimensional Feature Alignment (HDMFA) mechanism that facilitates the precise capture of multiple change cues at different levels through decoupled modeling of deep and shallow features and the design of multi-dimensional alignment branches. Additionally, we develop a Shallow Frequency-domain Feature Alignment (SFFA) branch that emphasizes the high-frequency components in the input data, effectively compensating for the inadequate representation of edge features in the spatial domain and enhancing the model's ability to detect complex boundaries. Experimental results on LEVIR-CD, SYSU-CD and EGY-BCD datasets validate the superiority of MDFANet, which outperforms SSLCD, Cmid and other SOTA methods by 1.7–2.3% in F1 score.

Recently, significant progress has been made in remote sensing image semantic segmentation based on attention and frequency. However, current attention mechanisms disregard the difference information between similar features, which reduces interclass variance and increases the difficulty of class discrimination. Meanwhile, the frequency domain’s limited representation of multidirectional edges and textures, coupled with weak feature complementarity in multiscale architectures, respectively, hinders contour capture and spatial localization of small objects. To address these issues, we propose a difference-guided multidirectional complementary network (DGMDC-Net). Specifically, a difference attention-guided aggregation module is constructed, which amplifies the differences between similar features in the aggregated information and serves as a guidance signal to reconstruct the response distribution of the global attention map, thereby facilitating category discrimination ability. Furthermore, we design a multidirectional frequency enhancement module, which amplifies the multidirectional characteristics of pixel intensity mutations in wavelet domain while supplementing the structural patterns of intensity distribution within local regions to refine the contour of small objects. In addition, a multilevel complementary module is developed, which correct detail deviations in deep layers while maintaining shallow contextual scene perception, to more accurately localize small targets. Experimental results on four benchmark datasets demonstrate that DGMDC-Net achieves superior performance in remote sensing image segmentation tasks.

Accurate large-scale crop mapping is critical for yield prediction, agricultural disaster monitoring, and global food security. Synthetic aperture radar (SAR), with its all-weather imaging capability, plays a vital role in remote sensing based on crop mapping studies. However, although feature-level fusion has been widely explored in remote sensing, existing VV and VH fusion approaches for rice mapping are still predominantly conducted at the data level and fail to adequately integrate their complementary information across the rice growth cycle, so the simplistic fusion methods yield features that are redundant or conflicting at field boundaries and in heterogeneous areas, thereby increasing classification errors. To address these challenges, this study proposes a novel spatiotemporal attention model (SimTA) for feature fusion to improve rice mapping. (1) A VV-VH feature-level fusion scheme is designed, integrated with a Content-Guided Attention (CGA) fusion method which effectively exploits the complementary information of the dual-polarized SAR data for achieving deep spatiotemporal dynamics fusion. (2) A Central Difference Convolution Spatial Extraction Conv (CDCSE Conv) Block is designed, enhancing sensitivity to edge variations in rice fields by combining standard and central difference convolutions. (3) To achieve efficient spatiotemporal feature integration across SAR time series, a Temporal–Spatial Attention (TSA) Block is developed, utilizing large-kernel convolutions for spatial feature extraction and a squeeze-and-excitation mechanism for capturing long-range temporal dependencies of rice time series. Extensive experiments were conducted by comparing SimTA with different models under five fusion schemes. Results demonstrate that feature-level fusion consistently outperforms other schemes, with SimTA achieving the best performance: OA = 91.1%, F1 score = 90.9%, and mIoU = 86.2%. Compared to the baseline Simple Video Prediction (SimVP), SimTA improves F1 score and mIoU by 0.8% and 2.1%, respectively. The CGA enhanced feature-level fusion further boosts SimTA’s performance to OA = 91.5% and F1 = 91.4%. SimTA bridges the gap between existing VV-VH deep fusion schemes and modern spatiotemporal modeling demands, offering a more accurate and generalizable approach for large-scale rice field mapping.

The invasion of alien plants has resulted in detrimental impactson the ecological, economic, and societal aspects.

Air temperature is a fundamental indicator for climate monitoring, agricultural planning, and ecosystem management. Land Surface Temperature (LST) retrieved from thermal remote sensing is widely used as a critical proxy for air temperature. However, existing air temperature estimation studies predominantly rely on single-date LST and diurnal variations (e.g., the Diurnal Temperature Cycle, DTC), overlooking the potential of multi-day LST temporal dependencies and lacking explicit spatial representativeness assessments. To address these challenges, here we present a hybrid Transformer-CNN daily mean air temperature (Ta) estimation network (HyTC-TaNet) which integrates Transformer-based temporal attention with convolutional feature extraction. While the Area of Applicability (AOA) metric is introduced to quantify spatial prediction confidence. Comparative experiments involving eight models, including HyTC-TaNet and seven benchmark algorithms, reveal that a 6-7 days temporal input window yields the best performance across all models. HyTC-TaNet achieves the highest accuracy, with RMSE = 1.429 dddddddddddddddddddddddddddddC, MAE = 1.101 ddddddddddddddddddddddddddddddddddC, and R222222222222222222222222 = 0.976, reducing errors by 0.273 dddddddddddddddddddddddC (RMSE) and 0.219 dddddddddddddddddddddddddddddddddddddddC (MAE) compared with concurrent day input. Spatial-temporal analysis further confirms that the optimal temporal sequence enriches historical and trend information, significantly enhancing estimation stability and spatial detail. Furthermore, the AOA of the model was assessed, confirming robust applicability in station-dense plains while effectively flagging extrapolation risks over large water bodies and high-altitude ridges due to limited representativeness in the training data. These findings demonstrate the strong potential of coupling HyTC-TaNet with optimally selected multi-day time series and spatial applicability assessment for precise, scalable air temperature estimation.