The performance and efficiency of small object detection are still very unsatisfactory due to the complex background interference for remote sensing images (RSIs) and the scale diversity among small objects. To solve this problem, this article proposes a new method, named “dynamic multi-scale enhanced feature fusion network (briefly, DMEFF-Net)” for small object detection in RSIs. The proposed framework integrates four key components: a multiscale convolutional decoupling module, which enhances feature representation by integrating multiple depth separable convolution layers with different dilation rates; a novel neck structure, the dynamic multiscale feature fusion network, which is built by integrating the fine-grained detail aggregation module to effectively represent feature information across different scales; a residual spatial context awareness module, which captures the relationship between local and global features to preserve details and understand full image context; and an adaptive threshold focal loss, which enhances small object detection by decoupling them from the dominant background. The model is evaluated on four public datasets, USOD, AI-TOD, NWPU-VHR 10, and RS-STOD, and compared with several state-of-the-art methods, including RefineDet, TPH-YOLOv5, SP-YOLOv8s, and FFCA-YOLO. The experimental results demonstrate that the proposed method achieves superior detection accuracy compared with existing approaches across all evaluated datasets.

Multiple-object tracking (MOT) is a fundamental task in computer vision and is widely applied across various domains. However, its algorithms remain somewhat immature in practical applications. To address the challenges presented by complex scenarios featuring instances of missed detections, false alarms, and frequent target switching leading to tracking failures, we propose an approach to multi-object tracking utilizing KC-YOLO detection and an identity validity discrimination module. We have constructed the KC-YOLO detection model as the detector for the tracking task, optimized the selection of detection frames, and implemented adaptive feature refinement to effectively address issues such as incomplete pedestrian features caused by occlusion. Furthermore, we have introduced an identity validity discrimination module in the data association component of the tracker. This module leverages the occlusion ratio coefficient, denoted by “k”, to assess the validity of pedestrian identities in low-scoring detection frames following cascade matching. This approach not only enhances pedestrian tracking accuracy but also ensures the integrity of pedestrian identities. In experiments on the MOT16, MOT17, and MOT20 datasets, MOTA reached 75.9%, 78.5%, and 70.1%, and IDF1 reached 74.8%, 77.8%, and 72.4%. The experimental results demonstrate the superiority of the methodology. This research outcome has potential applications in security monitoring, including public safety and fire prevention, for tracking critical targets.

With the increasing demand for remote sensing image applications, extracting the required images from a huge set of remote sensing images has become a hot topic. The previous retrieval methods cannot guarantee the efficiency, accuracy, and interpretability in the retrieval process. Therefore, we propose a bag-of-words association mapping method that can explain the semantic derivation process of remote sensing images. The method constructs associations between low-level features and high-level semantics through visual feature word packets. An improved FP-Growth method is proposed to achieve the construction of strong association rules to semantics. A feedback mechanism is established to improve the accuracy of subsequent retrievals by reducing the semantic probability of incorrect retrieval results. The public datasets AID and NWPU-RESISC45 were used to validate these experiments. The experimental results show that the average accuracies of the two datasets reach 87.5% and 90.8%, which are 22.5% and 20.3% higher than VGG16, and 17.6% and 15.6% higher than ResNet18, respectively. The experimental results were able to validate the effectiveness of our proposed method.

Traditional night light images are black and white with a low resolution, which has largely limited their applications in areas such as high-accuracy urban electricity consumption estimation. For this reason, this study proposes a fusion algorithm based on a dual-transformation (wavelet transform and IHS (Intensity Hue Saturation) color space transform), is proposed to generate color night light remote sensing images (color-NLRSIs). In the dual-transformation, the red and green bands of Landsat multi-spectral images and “NPP-VIIRS-like” night light remote sensing images are merged. The three bands of the multi-band image are converted into independent components by the IHS modulated wavelet transformed algorithm, which represents the main effective information of the original image. With the color space transformation of the original image to the IHS color space, the components $I$, $H$, and $S$ of Landsat multi-spectral images are obtained, and the histogram is optimally matched, and then it is combined with a two-dimensional discrete wavelet transform. Finally, it is inverted into RGB (red, green, and blue) color images. The experimental results demonstrate the following: (1) Compared with the traditional single-fusion algorithm, the dual-transformation has the best comprehensive performance effect on the spatial resolution, detail contrast, and color information before and after fusion, so the fusion image quality is the best; (2) The fused color-NLRSIs can visualize the information of the features covered by lights at night, and the resolution of the image has been improved from 500 m to 40 m, which can more accurately analyze the light of small-scale area and the ground features covered; (3) The fused color-NLRSIs are improved in terms of their MEAN (mean value), STD (standard deviation), EN (entropy), and AG (average gradient) so that the images have better advantages in terms of detail texture, spectral characteristics, and clarity of the images. In summary, the dual-transformation algorithm has the best overall performance and the highest quality of fused color-NLRSIs.

Night-light remote sensing imagery (NLRSI) effectively reflects urban economic and human activities and has important value in the field of remote sensing. However, the applications are limited by their low spatial resolution. In recent years, multisource remote sensing image fusion has become an important method for enhancing the spatial and spectral resolution of single-image data. To address the low-resolution limitation of NLRSI, this study proposes a multisource remote sensing image fusion framework based on the two-branch convolutional neural network (TbCNN), which fuses Landsat-8 and NPP/VIIRS data to generate high-resolution color night-light remote sensing imagery (CNLRSI). First, TbCNN features a deep two-branch structure for multiscale feature extraction, yielding richer spatial texture features. The framework also integrates a multilevel feature fusion module and a residual learning mechanism, further improving the fusion performance of CNLRSI. Quantitative evaluations demonstrate TbCNNs superiority over other methods, achieving optimal values in objective evaluation metrics. Second, in the built-up area extraction experiment, CNLRSI better identifies the true morphology of urban built-up areas compared with NPP/VIIRS data, reducing the overestimation of central urban areas and the underestimation of suburban areas in NPP/VIIRS data. Finally, the enhanced classification capability of CNLRSI is quantitatively validated through confusion matrix analysis, achieving a higher Kappa coefficient (0.814 versus 0.774) than NPP/VIIRS in urban pixel recognition.

With the development of various location-based social networks (LSBNs), personalized point-of-interest (POI) recommendations have become a recent research hotspot. Current recommendation methods tend to mine user preferences from their historical check-in records but overlook interest deviations caused by real-time geographic environments and immediate interests present in the records, failing to meet users’ real-time and accurate needs. Therefore, this paper proposes a composite preference-based recommendation model (CPRM) for personalized POI recommendation. This method first extracts multi-factor contextual features, constructs a dual-layer attention network (DLAN) to capture long and short-term preferences, combines real-time geographic scenarios to uncover user immediate preferences, and then weights and fuses these three types of preferences to generate user composite preferences. Finally, a prediction function is employed to obtain the Top-N recommendation list. The experiments on two classic datasets, Foursquare and Gowalla, affirm the effectiveness of the model presented in this paper and offer a novel approach for providing personalized POI recommendations to users.