In the wake of developments in remote sensing, the application of target detection of remote sensing is of increasing interest. Unfortunately, unlike natural image processing, remote sensing image processing involves dealing with large variations in object size, which poses a great challenge to researchers. Although traditional multi-scale detection networks have been successful in solving problems with such large variations, they still have certain limitations: (1) The traditional multi-scale detection methods note the scale of features but ignore the correlation between feature levels. Each feature map is represented by a single layer of the backbone network, and the extracted features are not comprehensive enough. For example, the SSD network uses the features extracted from the backbone network at different scales directly for detection, resulting in the loss of a large amount of contextual information. (2) These methods combine with inherent backbone classification networks to perform detection tasks. RetinaNet is just a combination of the ResNet-101 classification network and FPN network to perform the detection tasks; however, there are differences in object classification and detection tasks. To address these issues, a cross-scale feature fusion pyramid network (CF2PN) is proposed. First and foremost, a cross-scale fusion module (CSFM) is introduced to extract sufficiently comprehensive semantic information from features for performing multi-scale fusion. Moreover, a feature pyramid for target detection utilizing thinning U-shaped modules (TUMs) performs the multi-level fusion of the features. Eventually, a focal loss in the prediction section is used to control the large number of negative samples generated during the feature fusion process. The new architecture of the network proposed in this paper is verified by DIOR and RSOD dataset. The experimental results show that the performance of this method is improved by 2–12% in the DIOR dataset and RSOD dataset compared with the current SOTA target detection methods.

Recently, deep learning methods based on the combination of spatial and spectral features have been successfully applied in hyperspectral image (HSI) classification. To improve the utilization of the spatial and spectral information from the HSI, this paper proposes a unified network framework using a three-dimensional convolutional neural network (3-D CNN) and a band grouping-based bidirectional long short-term memory (Bi-LSTM) network for HSI classification. In the framework, extracting spectral features is regarded as a procedure of processing sequence data, and the Bi-LSTM network acts as the spectral feature extractor of the unified network to fully exploit the close relationships between spectral bands. The 3-D CNN has a unique advantage in processing the 3-D data; therefore, it is used as the spatial-spectral feature extractor in this unified network. Finally, in order to optimize the parameters of both feature extractors simultaneously, the Bi-LSTM and 3-D CNN share a loss function to form a unified network. To evaluate the performance of the proposed framework, three datasets were tested for HSI classification. The results demonstrate that the performance of the proposed method is better than the current state-of-the-art HSI classification methods.

Most current research primarily focuses on improving experimental accuracy using large models, often neglecting the deployment challenges. There is a growing need for lightweight algorithms in certain remote sensing devices. Moreover, remote sensing images (RSIs) often contain numerous small, densely distributed targets, which pose significant detection challenges. To address these issues, we have improved the YOLOv8s network and developed a lightweight remote sensing object detection (RSOD) network based on multiscale features fusion and context information (MFFCI-YOLOv8). This network combines multiscale feature fusion and contextual information to accurately detect objects in RSIs. Firstly, we introduce the lightweight CSP bottleneck with attention (LCA) module, which utilizes partial convolution calculation and SimAM attention mechanisms to decrease the number of param-eters and computational complexity while enhancing feature extraction capabilities. Secondly, we design the gate spatial pyramid pooling fast (GSPPF) module to enhance the model's perception of scale and contextual information, thus improving the detection of small objects. Lastly, we employ the multiscale fusion lightweight neck (MFLNeck) module for more efficient multiscale feature fusion, preventing the loss of small objects. Compared to YOLOv8s, our overall model reduces the number of parameters by 7.7% and FLOPs by 11.9%. We validated the accuracy of MFFCI-YOLOv8 on two remote sensing datasets, NWPU VHR-10 and VisDrone. The experimental results demons-trate that our model offers a low computational cost and high detection accuracy compared to other RSOD models and other YOLO models.

Currently, convolutional neural network (CNN) and transformer-based hyperspectral image (HSI) classification methods have attracted significant attention owing to their effective feature representation capabilities. However, methods based on CNN pay insufficient attention to valuable pixels in 3D HSI samples and cannot adapt to variations in these samples. Transformer-based methods also suffer from high computational complexity and a tendency for low-level spatial-spectral features of the shallow attention layer to vanish as the number of attention layers increases. To address these issues, we proposed a local–global feature extraction network with dynamic 3D convolution and residual attention transformer (LGDRNet). The LGDRNet primarily consists of multiscale 3D conv, dynamic local feature extraction, residual global feature extraction, and feature fusion modules. Specifically, a multiscale 3D conv module is used for low-level multiscale spectral information extraction. Then, the dynamic local feature extraction module utilizes dynamic 3D convolution, which can adapt to different samples. This allows the network to focus on valuable pixels in 3D samples. The residual global feature extraction module utilizes a convolutional projection unit and convolutional multihead self-attention to reduce computational complexity. It employs a residual attention connection to enable the network to effectively transmit and accumulate attention information across consecutive multihead attention layers. This prevents the vanishing of shallow spatial–spectral features. Finally, local and global HSI information may be efficiently integrated using the feature fusion module, which also improves performance during subsequent classification. The proposed model achieves overall classification accuracies of 89.24%, 92.01%, and 94.53% on three benchmark datasets, respectively, outperforming state-of-the-art approaches with limited training samples

The combination of Convolutional neural networks (CNNs) and Vision transformers (ViTs) has garnered considerable attention in hyperspectral image (HSI) classification due to their abilities to enhance the classification accuracy by concurrently extracting local and global features. However, these accuracy improvements come at the cost of significant demands on storage resources, computational overhead, and extensive training samples. To address these challenges, this paper proposes a Multi-scale spatial-spectral shuffling convolution integrated with a three-dimensional lightweight transformer (MSC-3DLT) for HSI classification. This network directly captures three-dimensional structural features throughout the entire feature extraction process, thereby enhancing HSI classification performance even at small sampling rates within a lightweight framework. Specifically, we design a multi-scale spatial-spectral shuffling convolution (MSC) to comprehensively refines spatial-spectral feature granularities and enhance feature interactions by shuffling multi-scale features across different groups. Second, to maximize the exploitation of limited training samples, we rethink transformers from the three-dimensional structural perspective of HSI data and propose a novel lightweight three-dimensional transformer (3DLT). Different from the slicing operation employed in classical transformers, the 3DLT directly extracts the inherent three-dimensional structural features from HSI and mitigates quadratic complexity through a lightweight spatial-spectral pooling cross-attention mechanism (LSPCA). Finally, a novel training strategy is designed to adaptively adjust the learning rate based on multi-metric feedback during the model training process, significantly accelerating the model fitting speed. Extensive experiments demonstrate that the proposed MSC-3DLT method remains highly competitive compared to state-of-the-art methods in terms classification accuracy, model parameters, and FLOPs under small sampling rates. The code of this work is available at https://github.com/H892328874/MSC-3DLT for the sake of reproducibility.

Convolutional neural networks (CNNs) and graph convolutional networks (GCNs) have led to promising advancements in hyperspectral image (HSI) classification; however, traditional CNNs with fixed square convolution kernels are insufficiently flexible to handle irregular structures. Similarly, GCNs that employ superpixel nodes instead of pixel nodes may overlook pixel-level features; both networks tend to extract features locally and cause loss of multilayer contextual semantic information during feature extraction due to the fixed kernel. To leverage the strengths of CNNs and GCNs, we propose a multiscale pixel-level and superpixel-level (MPAS)-based HSI classification method. The network consists of two sub-networks for extracting multi-level information of HSIs: a multi-scale hybrid spectral–spatial attention convolution branch (HSSAC) and a parallel multi-hop graph convolution branch (MGCN). HSSAC comprehensively captures pixel-level features with different kernel sizes through parallel multi-scale convolution and cross-path fusion to reduce the semantic information loss caused by fixed convolution kernels during feature extraction and learns adjustable weights from the adaptive spectral–spatial attention module (SSAM) to capture pixel-level feature correlations with less computation. MGCN can systematically aggregate multi-hop contextual information to better model HSIs’ spatial background structure using the relationship between parallel multi-hop graph transformation nodes. The proposed MPAS effectively captures multi-layer contextual semantic features by leveraging pixel-level and superpixel-level spectral–spatial information, which improves the performance of the HSI classification task while ensuring computational efficiency. Extensive evaluation experiments on three real-world HSI datasets demonstrate that MPAS outperforms other state-of-the-art networks, demonstrating its superior feature learning capabilities.