Multi-scale Features Research Articles

Real-Time Detection and Counting of Wheat Spikes Based on Improved YOLOv10

Wheat is one of the most crucial food crops globally, with its yield directly impacting global food security. The accurate detection and counting of wheat spikes is essential for monitoring wheat growth, predicting yield, and managing fields. However, the current methods face challenges, such as spike size variation, shading, weed interference, and dense distribution. Conventional machine learning approaches have partially addressed these challenges, yet they are hampered by limited detection accuracy, complexities in feature extraction, and poor robustness under complex field conditions. In this paper, we propose an improved YOLOv10 algorithm that significantly enhances the model’s feature extraction and detection capabilities. This is achieved by introducing a bidirectional feature pyramid network (BiFPN), a separated and enhancement attention module (SEAM), and a global context network (GCNet). BiFPN leverages both top-down and bottom-up bidirectional paths to achieve multi-scale feature fusion, improving performance in detecting targets of various scales. SEAM enhances feature representation quality and model performance in complex environments by separately augmenting the attention mechanism for channel and spatial features. GCNet captures long-range dependencies in the image through the global context block, enabling the model to process complex information more accurately. The experimental results demonstrate that our method achieved a precision of 93.69%, a recall of 91.70%, and a mean average precision (mAP) of 95.10% in wheat spike detection, outperforming the benchmark YOLOv10 model by 2.02% in precision, 2.92% in recall, and 1.56% in mAP. Additionally, the coefficient of determination (R2) between the detected and manually counted wheat spikes was 0.96, with a mean absolute error (MAE) of 3.57 and a root-mean-square error (RMSE) of 4.09, indicating strong correlation and high accuracy. The improved YOLOv10 algorithm effectively solves the difficult problem of wheat spike detection under complex field conditions, providing strong support for agricultural production and research.

Multi-Scale Depthwise Separable Capsule Network for hyperspectral image classification.

Addressing the challenges in effectively extracting multi-scale features and preserving pose information during hyperspectral image (HSI) classification, a Multi-Scale Depthwise Separable Capsule Network (MDSC-Net) is proposed in this article for HSI classification. Initially, hierarchical features are extracted by MDSC-Net through the employment of parallel multi-scale convolutional kernels, while computational complexity is reduced via depthwise separable convolutions, thus reducing the overall computational load and achieving efficient feature extraction. Subsequently, to enhance the translational invariance of features and reduce the loss of pose information, features of various scales are processed in parallel by independent capsule networks, with improvements in max pooling achieved through dynamic routing. Lastly, features of different scales are concatenated and integrated through the concatenate operation, thereby facilitating precise analysis of multi-level information in the hyperspectral image classification process. Experimental comparisons demonstrate that MDSC-Net achieves average accuracies of 94%, 98%, and 99% on the Kennedy Space Center, University of Pavia, and Salinas datasets, respectively, indicating a significant performance advantage over recent HSI classification models and validating the effectiveness of the proposed model.

MaskRecon: High-quality human reconstruction via masked autoencoders using a single RGB-D image

In this paper, we explore reconstructing high-quality clothed 3D humans from a single RGB-D image, assuming that virtual humans can be represented by front-view and back-view depths. Due to the scarcity of captured real RGB-D human images, we employ rendered images to train our method. However, rendered images lack background with significant depth variation in silhouettes, leading to shape prediction inaccuracies and noise. To mitigate this issue, we introduce a pseudo-multi-task framework, which incorporates a Conditional Generative Adversarial Network (CGAN) to infer back-view RGB-D images and a self-supervised Masked Autoencoder (MAE) to capture latent structural information of the human body. Additionally, we propose a Multi-scale Feature Fusion (MFF) module to effectively merge structural information and conditional features at various scales. Our method surpasses many existing techniques, as demonstrated through evaluations on the Thuman, RenderPeople, and BUFF datasets. Notably, our approach excels in reconstructing high-quality human models, even under challenging conditions such as complex poses and loose clothing, both on rendered and real-world images. Codes are available at https://github.com/Archaic-Atom/MaskRecon.

Generating crisp boundaries using multi-scale features and mixed loss function

Generating crisp boundaries using multi-scale features and mixed loss function

Dual-consistency guidance semi-supervised medical image segmentation with low-level detail feature augmentation

In deep-learning-based medical image segmentation tasks, semi-supervised learning can greatly reduce the dependence of the model on labeled data. However, existing semi-supervised medical image segmentation methods face the challenges of object boundary ambiguity and a small amount of available data, which limit the application of segmentation models in clinical practice. To solve these problems, we propose a novel semi-supervised medical image segmentation network based on dual-consistency guidance, which can extract reliable semantic information from unlabeled data over a large spatial and dimensional range in a simple and effective manner. This serves to improve the contribution of unlabeled data to the model accuracy. Specifically, we construct a split weak and strong consistency constraint strategy to capture data-level and feature-level consistencies from unlabeled data to improve the learning efficiency of the model. Furthermore, we design a simple multi-scale low-level detail feature enhancement module to improve the extraction of low-level detail contextual information, which is crucial to accurately locate object contours and avoid omitting small objects in semi-supervised medical image dense prediction tasks. Quantitative and qualitative evaluations on six challenging datasets demonstrate that our model outperforms other semi-supervised segmentation models in terms of segmentation accuracy and presents advantages in terms of generalizability. Code is available at https://github.com/0Jmyy0/SSMIS-DC.

An Intelligent Fishery Detection Method Based on Cross-Domain Image Feature Fusion

Target detection technology plays a crucial role in fishery ecological monitoring, fishery diversity research, and intelligent aquaculture. Deep learning, with its distinct advantages, provides significant convenience to the fishery industry. However, it still faces various challenges in practical applications, such as significant differences in image species and image blurring. To address these issues, this study proposes a multi-scale, multi-level, and multi-stage cross-domain feature fusion model. In order to train the model more effectively, a new data set called Fish52 (multi-scene fish data set, a data set containing 52 fish species) was constructed, on which the model achieved an mAP (mean average precision is a key measure of model performance) of 82.57%. Furthermore, we compared prevalent one-stage and two-stage detection methods on the Lahatan (single-scene fish data set) and Fish30 data set (a data set containing 30 fish species) and tested them on the F4k (Fish4Knowledge (F4K) is a data set focused on fish detection and identification) and FishNet data set (it is a data set containing 94,532 images from 17,357 aquatic species). The mAP of our proposed model on the Fish30, Lahatan, F4k, and FishNet data sets reaches 91.72%, 98.7%, 88.6%, and 81.5%, respectively, outperforming existing mainstream models. Comprehensive empirical analysis indicates that our model possesses a high generalization ability and reaches advanced performance levels. In this study, the depth of the model backbone is deepened, a novel neck structure is proposed, and a new module is embedded therein. To enhance the fusion ability of the model, a new attention mechanism module is introduced. In addition, in the adaptive decoupling detection head module, introducing classes with independent parameters and regression adapters reduces interaction between different tasks. The proposed model can better monitor fishery resources and enhance aquaculture efficiency. It not only provides an effective approach for fish detection but also has certain reference significance for the identification of similar targets in other environments and offers assistance for the construction of smart fisheries and digital fisheries.

Cross-Hopping Graph Networks for Hyperspectral–High Spatial Resolution (H2) Image Classification

As we take stock of the contemporary issue, remote sensing images are gradually advancing towards hyperspectral–high spatial resolution (H2) double-high images. However, high resolution produces serious spatial heterogeneity and spectral variability while improving image resolution, which increases the difficulty of feature recognition. So as to make the best of spectral and spatial features under an insufficient number of marking samples, we would like to achieve effective recognition and accurate classification of features in H2 images. In this paper, a cross-hop graph network for H2 image classification(H2-CHGN) is proposed. It is a two-branch network for deep feature extraction geared towards H2 images, consisting of a cross-hop graph attention network (CGAT) and a multiscale convolutional neural network (MCNN): the CGAT branch utilizes the superpixel information of H2 images to filter samples with high spatial relevance and designate them as the samples to be classified, then utilizes the cross-hop graph and attention mechanism to broaden the range of graph convolution to obtain more representative global features. As another branch, the MCNN uses dual convolutional kernels to extract features and fuse them at various scales while attaining pixel-level multi-scale local features by parallel cross connecting. Finally, the dual-channel attention mechanism is utilized for fusion to make image elements more prominent. This experiment on the classical dataset (Pavia University) and double-high (H2) datasets (WHU-Hi-LongKou and WHU-Hi-HongHu) shows that the H2-CHGN can be efficiently and competently used in H2 image classification. In detail, experimental results showcase superior performance, outpacing state-of-the-art methods by 0.75–2.16% in overall accuracy.

Enhanced 3D object detection for autonomous driving: A spatial-temporal alignment approach in Bird's Eye View scenarios

This paper presents a novel 3D object detection algorithm designed for Bird's Eye View (BEV) scenarios, which significantly improves detection capabilities by integrating spatial and temporal features. The core of our approach is the spatial-temporal alignment module that efficiently processes information across different time steps and spatial locations, enhancing the precision and robustness of object detection. We employ a temporal self-attention mechanism to capture the motion information of objects over time, allowing the model to correlate features across various time steps for identifying and tracking moving objects. Additionally, a spatial cross-attention mechanism is utilized to focus on spatial features within regions of interest, promoting interactions between features extracted from camera views and BEV queries. Our method also implements temporal feature integration and multi-scale feature fusion to enhance detection stability and accuracy for fast-moving objects and to capture multi-scale context information, respectively. The model employs an enriched feature set post alignment for 3D bounding box prediction, ascertaining the position, dimensions, and orientation of objects. We conducted experiments on two public datasets for autonomous driving nuScenes and Waymo Open Dataset, demonstrating that our method outperforms previous BEVFormer and other state-of-the-art methods in terms of detection accuracy and robustness. The paper concludes with potential future directions for optimizing the BEVFormer model's performance and exploring its application in broader scenarios and tasks.

Using unmanned aerial vehicle acquired RGB images and Density-Cluster-Count model for tree-level apple flower quantification

Accurate estimation of apple flower quantity is vital for flower thinning strategies, yield prediction, and other aspects related to orchard management. Compared to flower quantity estimation based on partial regions, achieving tree-level flower quantification is more meaningful. To achieve tree-level apple flower cluster counting, an aerial apple tree image-based approach was developed based on density estimation and density peak. Initially, images of apple trees captured by UAV were input into the apple flower density estimation model, which was based on a multi-scale feature fusion network, to obtain a density map of apple flowers. Utilizing the growth patterns and distribution characteristics of apple flowers and clusters, apple flower cluster counting was accomplished through the identification of local density peaks. Following testing, the proposed method achieved a Mean Absolute Error (MAE) of 5.39 and a Root Mean Squared Error (RMSE) of 8.10 in apple flower cluster counting. The MAE was decreased by 3.83, 16.22, and 11.13 compared to apple flower cluster counting methods based on YOLOv5s, U-Net, and MCNN, respectively. Additionally, the RMSE was 4.44, 22.13, and 15.79 lower compared to the three aforementioned methods. This research facilitated a relatively accurate counting of apple flower clusters at the tree-level in orchards, offering valuable information for tasks including flowering monitoring, load estimation, and yield forecasting.

PCB plug-in solder joint defect detection method based on coordinated attention-guided information fusion

Printed Circuit Boards (PCBs) are the foundational component of electronic devices, and the detection of PCB defects is essential for ensuring the quality control of electronic products. Aiming at the problem that the existing PCB plug-in solder defect detection algorithms cannot meet the requirements of high precision, low false alarm rate, and high speed at the same time, this paper proposes a method based on spatial convolution pooling and information fusion. Firstly, on the basis of YOLOv3, an attention-guided pyramid structure is used to fuse context information, and multiple convolutions of different size are used to explore richer high-level semantic information; Secondly, a coordinated attention network structure is introduced to calibrate the fused pyramidal feature information, highlighting the important feature channels, and reducing the adverse impact of redundant parameters generated by feature fusion; Finally, the ASPP (Atrous Spatial Pyramid Pooling) structure is implemented in the original Darknet53 backbone feature extraction network to acquire multi-scale feature information of the detection targets. With these improvements, the average detection accuracy of the enhanced network has been elevated from 94.45 to 96.43%. This experiments shows that the improved network is more suitable for PCB plug-in solder defect detection applications.

WaveFusionNet: Infrared and visible image fusion based on multi-scale feature encoder–decoder and discrete wavelet decomposition

To merge complementary information from multimodal images, such as thermal saliency from infrared images and texture details from visible images, traditional multi-scale transform-based methods have been extensively studied, with deep learning-based methods gaining significant popularity in recent years. However, there has been limited research on optimally combining the advantages of these two categories in fusion. In this paper, we propose a novel infrared and visible image fusion (IVIF) framework, WaveFusionNet, which integrates precise frequency feature decomposition from the discrete wavelet transform (DWT) with the comprehensive feature extraction from the multi-scale encoder. Firstly, we train an encoder–decoder network for multi-scale feature extraction and image reconstruction. DWT is used for down-sampling with minimal information loss by decomposing extracted features into low and high-frequency sub-bands. Next, a dual-band feature fusion (DBFF) module is trained to merge these sub-bands by integrating a spatial feature transform-based sub-network for low-frequency fusion and a maximum absolute value selection strategy for fusing high-frequencies. Finally, all fused sub-bands are fed into the pre-trained decoder to reconstruct the final image. Experimental results on three benchmark datasets (TNO, Roadscene, and MSRS) demonstrate that the proposed fusion method outperforms recent IVIF methods in both quantitative assessment and visual perception while maintaining competitive time complexity.

Using the MSFNet Model to Explore the Temporal and Spatial Evolution of Crop Planting Area and Increase Its Contribution to the Application of UAV Remote Sensing

Remote sensing technology can be used to monitor changes in crop planting areas to guide agricultural production management and help achieve regional carbon neutrality. Agricultural UAV remote sensing technology is efficient, accurate, and flexible, which can quickly collect and transmit high-resolution data in real time to help precision agriculture management. It is widely used in crop monitoring, yield prediction, and irrigation management. However, the application of remote sensing technology faces challenges such as a high imbalance of land cover types, scarcity of labeled samples, and complex and changeable coverage types of long-term remote sensing images, which have brought great limitations to the monitoring of cultivated land cover changes. In order to solve the abovementioned problems, this paper proposed a multi-scale fusion network (MSFNet) model based on multi-scale input and feature fusion based on cultivated land time series images, and further combined MSFNet and Model Diagnostic Meta Learning (MAML) methods, using particle swarm optimization (PSO) to optimize the parameters of the neural network. The proposed method is applied to remote sensing of crops and tomatoes. The experimental results showed that the average accuracy, F1-score, and average IoU of the MSFNet model optimized by PSO + MAML (PSML) were 94.902%, 91.901%, and 90.557%, respectively. Compared with other schemes such as U-Net, PSPNet, and DeepLabv3+, this method has a better effect in solving the problem of complex ground objects and the scarcity of remote sensing image samples and provides technical support for the application of subsequent agricultural UAV remote sensing technology. The study found that the change in different crop planting areas was closely related to different climatic conditions and regional policies, which helps to guide the management of cultivated land use and provides technical support for the realization of regional carbon neutrality.

Small Object Detection in UAV Images Based on YOLOv8n

With the rapid development of unmanned aerial vehicle (UAV) technology, there is an urgent need for high-performance aerial object detection algorithms that are tailored for deployment on drones with limited computing capabilities. This paper proposes a series of improvements to the state-of-the-art YOLOv8 object detector to enhance its detection accuracy and speed for small, partially occluded objects in complex environments. Specifically, we introduce multi-scale feature fusion through additional detection layers, employ conditionally parameterized convolutions to increase representational capacity, and import a dynamic non-monotonic loss function named Wise-IoU to enable more effective regression of bounding boxes. Experiments conducted on the large-scale UAV benchmark dataset VisDrone demonstrate that the improved model achieves state-of-the-art accuracy of 37.6% mAP with 3 M parameters, outperforming other lightweight YOLO detectors and two-stage detectors like Faster R-CNN. The improved model also reaches 40 FPS on an embedded edge device, validating its efficiency and suitability for real-time UAV applications. Through comprehensive quantitative experiments and visual results, this work provides valuable insights and techniques to tailor object detection algorithms for robust and efficient deployment on UAVs with limited onboard computing power.

Language conditioned multi-scale visual attention networks for visual grounding

Visual grounding (VG) is a task that requires to locate a specific region in an image according to a natural language expression. Existing efforts on the VG task are divided into two-stage, one-stage and Transformer-based methods, which have achieved high performance. However, most of the previous methods extract visual information at a single spatial scale and ignore visual information at other spatial scales, which makes these models unable to fully utilize the visual information. Moreover, the insufficient utilization of linguistic information, especially failure to capture global linguistic information, may lead to failure to fully understand language expressions, thus limiting the performance of these models. To better address the task, we propose a language conditioned multi-scale visual attention network (LMSVA) for visual grounding, which can sufficiently utilize visual and linguistic information to perform multimodal reasoning, thus improving performance of model. Specifically, we design a visual feature extractor containing a multi-scale layer to get the required multi-scale visual features by expanding the original backbone. Moreover, we exploit pooling the output of the pre-trained Bidirectional Encoder Representations from Transformers (BERT) model to extract sentence-level linguistic features, which can enable the model to capture global linguistic information. Inspired by the Transformer architecture, we present the Visual Attention Layer guided by Language and Multi-Scale Visual Features (VALMS), which is able to better learn the visual context guided by multi-scale visual and linguistic features, and facilitates further multimodal reasoning. Extensive experiments on four large benchmark datasets, including ReferItGame, RefCOCO, RefCOCO+ and RefCOCOg, demonstrate that our proposed model achieves the state-of-the-art performance.

Stepwise Attention-Guided Multiscale Fusion Network for Lightweight and High-Accurate SAR Ship Detection

Many exceptional deep learning networks have demonstrated remarkable proficiency in general object detection tasks. However, the challenge of detecting ships in synthetic aperture radar (SAR) imagery increases due to the complex and various nature of these scenes. Moreover, sophisticated large-scale models necessitate substantial computational resources and hardware expenses. To address these issues, a new framework is proposed called a stepwise attention-guided multiscale feature fusion network (SAFN). Specifically, we introduce a stepwise attention mechanism designed to selectively emphasize relevant information and filter out irrelevant details of objects in a step-by-step manner. Firstly, a novel LGA-FasterNet is proposed, which incorporates a lightweight backbone FasterNet with lightweight global attention (LGA) to realize expressive feature extraction while reducing the model’s parameters. To effectively mitigate the impact of scale and complex background variations, a deformable attention bidirectional fusion network (DA-BFNet) is proposed, which introduces a novel deformable location attention (DLA) block and a novel deformable recognition attention (DRA) block, strategically integrating through bidirectional connections to achieve enhanced features fusion. Finally, we have substantiated the robustness of the new framework through extensive testing on the publicly accessible SAR datasets, HRSID and SSDD. The experimental outcomes demonstrate the competitive performance of our approach, showing a significant enhancement in ship detection accuracy compared to some state-of-the-art methods.

BCMFIFuse: A Bilateral Cross-Modal Feature Interaction-Based Network for Infrared and Visible Image Fusion

The main purpose of infrared and visible image fusion is to produce a fusion image that incorporates less redundant information while incorporating more complementary information, thereby facilitating subsequent high-level visual tasks. However, obtaining complementary information from different modalities of images is a challenge. Existing fusion methods often consider only relevance and neglect the complementarity of different modalities’ features, leading to the loss of some cross-modal complementary information. To enhance complementary information, it is believed that more comprehensive cross-modal interactions should be provided. Therefore, a fusion network for infrared and visible fusion is proposed, which is based on bilateral cross-feature interaction, termed BCMFIFuse. To obtain features in images of different modalities, we devise a two-stream network. During the feature extraction, a cross-modal feature correction block (CMFC) is introduced, which calibrates the current modality features by leveraging feature correlations from different modalities in both spatial and channel dimensions. Then, a feature fusion block (FFB) is employed to effectively integrate cross-modal information. The FFB aims to explore and integrate the most discriminative features from the infrared and visible image, enabling long-range contextual interactions to enhance global cross-modal features. In addition, to extract more comprehensive multi-scale features, we develop a hybrid pyramid dilated convolution block (HPDCB). Comprehensive experiments on different datasets reveal that our method performs excellently in qualitative, quantitative, and object detection evaluations.

Gait recognition for farm workers via multi-scale temporal feature perception

The monitoring and management of farm workers play a crucial role in agricultural production. The real-time monitoring of farm workers can effectively identify them, facilitating standardized management and operation of the farm. Gait, as a unique biometric feature that can be recognized from a distance, is more suitable for identifying individual farm workers walking on the farm at a distance compared to facial and iris features. Despite the increasing research on gait recognition techniques, existing gait recognition models are still lacking in temporal modeling. The gait of farm workers walking through farmland is greatly influenced by the environment, making it crucial to model the temporal information of gait. In this study, we captured videos of 50 farm workers from 11 different perspectives walking in real agricultural fields and formed the Farm Worker Gait Dataset (FWGD). In comparison with existing gait recognition models, we present a multi-scale temporal feature modeling method based on the GaitBase model and obtain a performance improvement on the FWGD dataset. Specifically, we propose an Adaptive Temporal Self-Attention (ATSA) mechanism to extract the most representative features based on temporal contextual temporal relations. Additionally, we utilize a dilated convolutional layer to extract discrete temporal features and employ a Multi-scale Temporal Perception Module (MSTP) with three branches to perceive and aggregate temporal features. With the assistance of MSTP and ATSA, our model significantly improves the discriminability of temporal features, thus enhancing the performance of farm worker gait recognition. Extensive experiments demonstrated that our proposed method outperformed state-of-the-art gait recognition methods on the FWGD dataset.

Improved high dynamic range imaging using multi-scale feature flows balanced between task-orientedness and accuracy

Deep learning has made it possible to accurately generate high dynamic range (HDR) images from multiple images taken at different exposure settings, largely owing to advancements in neural network design. However, generating images without artifacts remains difficult, especially in scenes with moving objects. In such cases, issues like color distortion, geometric misalignment, or ghosting can appear. Current state-of-the-art network designs address this by estimating the optical flow between input images to align them better. The parameters for the flow estimation are learned through the primary goal, producing high-quality HDR images. However, we find that this ”task-oriented flow” approach has its drawbacks, especially in minimizing artifacts. To address this, we introduce a new network design and training method that improve the accuracy of flow estimation. This aims to strike a balance between task-oriented flow and accurate flow. Additionally, the network utilizes multi-scale features extracted from the input images for both flow estimation and HDR image reconstruction. Our experiments demonstrate that these two innovations result in HDR images with fewer artifacts and enhanced quality.

Fracture properties of green nano fibrous network with random and aligned fiber distribution: A hierarchical molecular dynamics and peridynamics approach

Polylactic acid (PLA) nanofibrous networks have gained substantial interest across various engineering and scientific disciplines, such as tissue engineering, drug delivery, and filtration, due to their unique and multifunctional attributes, including biodegradability, tuneable mechanical properties, and surface functionality. However, predicting their mechanical behaviour remains challenging due to their structural complexity, multiscale features, and variability in material properties.This study presents a hierarchical approach to investigate the fracture phenomena in both aligned and randomly oriented nanofibrous networks by integrating atomistic modelling and non-local continuum mechanics, peridynamics. At the nanoscale, all-atom molecular dynamics simulations are employed to apply tensile loads to freestanding pristine and silver-doped PLA nanofibres, where key mechanical properties such as Young's modulus, Poisson's ratio, and critical energy release rate are determined using innovative approaches. A new method is introduced to seamlessly transfer data from molecular dynamics to peridynamics by ensuring the convergence of the tensile response of a single fiber in both frameworks. This nano to micro coupling technique is then utilised to examine the Young's modulus, fracture toughness of mode I and II, and crack propagation in PLA nanofibrous networks. The proposed framework can also incorporate the effects of surface coating and fiber arrangements on the measured properties. The current research paves the way for the development of stronger and more durable eco-friendly nanofibrous networks with optimised performance.

MFCF-Gait: Small Silhouette-Sensitive Gait Recognition Algorithm Based on Multi-Scale Feature Cross-Fusion.

Gait recognition based on gait silhouette profiles is currently a major approach in the field of gait recognition. In previous studies, models typically used gait silhouette images sized at 64 × 64 pixels as input data. However, in practical applications, cases may arise where silhouette images are smaller than 64 × 64, leading to a loss in detail information and significantly affecting model accuracy. To address these challenges, we propose a gait recognition system named Multi-scale Feature Cross-Fusion Gait (MFCF-Gait). At the input stage of the model, we employ super-resolution algorithms to preprocess the data. During this process, we observed that different super-resolution algorithms applied to larger silhouette images also affect training outcomes. Improved super-resolution algorithms contribute to enhancing model performance. In terms of model architecture, we introduce a multi-scale feature cross-fusion network model. By integrating low-level feature information from higher-resolution images with high-level feature information from lower-resolution images, the model emphasizes smaller-scale details, thereby improving recognition accuracy for smaller silhouette images. The experimental results on the CASIA-B dataset demonstrate significant improvements. On 64 × 64 silhouette images, the accuracies for NM, BG, and CL states reached 96.49%, 91.42%, and 78.24%, respectively. On 32 × 32 silhouette images, the accuracies were 94.23%, 87.68%, and 71.57%, respectively, showing notable enhancements.