Multi-scale Channel Research Articles

Using convolutional neural network combined with multi-scale channel attention module to predict soil properties from visible and near-infrared spectral data

With the increasing demand for precision agriculture and sustainable land management, it is crucial to quickly and accurately predict soil properties. However, there are still challenges in predicting large-scale soil properties. This study proposes a new convolutional neural network architecture (MCA-CNN) combined with visible and near-infrared spectroscopy for predicting soil properties. The architecture introduces a multi-scale channel attention mechanism, aiming to more effectively capture complex features in the spectrum. The performance of the MCA-CNN model was tested on a large-scale dataset and compared with partial least squares (PLS), extreme gradient boosting tree (XGBoost), and conventional convolutional neural network (CNN) models. The predictive performance of different models in organic carbon (OC), nitrogen (N), pH, Clay, Salt, and Sand was analyzed. The results showed that the MCA-CNN model had the best predictive performance on all soil properties, achieving the lowest root mean square error (RMSE) and the highest coefficient of determination (R2), the R2 of OC, N, pH, Clay, Silt, and Sand are 0.95, 0.93, 0.95, 0.80, 0.58, 0.64, respectively. The RMSE of OC, N, pH, Clay, Silt, and Sand are 16.60 g kg−1, 0.96 g kg−1, 0.31, 4.89 %, 8.27 %, and 11.61 %, respectively. Further research has found that soil types (mineral soil and organic soil) affect the predictive performance of OC. It is necessary to distinguish between mineral soil and organic soil when predicting soil OC content. In addition, the effectiveness of the model transfer strategy was explored for situations with small soil sample sizes. By fine-tuning the pre-trained model, the performance of the model can be significantly improved, which provides a feasible solution for predicting soil properties in situations of data scarcity. In summary, this study not only confirms the potential application of deep learning in the field of soil science, but also provides new technological avenues for future soil management and agricultural practices.

Wideband Adaptive Beamforming via Multiscale Channel Attention Convolutional Neural Network

Wideband Adaptive Beamforming via Multiscale Channel Attention Convolutional Neural Network

Acoustic emission-based weld crack leakage monitoring via FGI and MCCF-CondenseNet convolutional neural network

Online monitoring of weld crack leakage in pressure pipelines of nuclear power ship based on acoustic emission (AE) technology is of great significance for maintaining the safe and stable operation of the system. However, most of the current leakage studies are conducted through artificially designed pipeline hole types, which deviate from the actual crack morphology and are weakly online, with low identification accuracy and slow monitoring speed. Therefore, a convolutional network of FGI and multi-scale channel information cross fusion based on AE technology is proposed in this paper. First, the FBank feature of the AE signal of pipeline weld leakage are extracted. On this basis, the Gini Index (GI) preference feature is used to filter the useless information in the FBank feature. Then, a multi-scale channel information cross fusion module is designed to improve the feature learning ability of the network through the interaction and fusion of different channel information. Finally, the superiority of the proposed FGI feature extraction method and the effectiveness of the proposed multi-scale channel information cross fusion CondenseNet (MCCF-CondenseNet) convolutional neural network are verified by the pipeline leakage AE monitoring experiments under three crack morphologies. The results show that the identification accuracy of the proposed method is as high as 96.42 %, and the identification speed is significantly faster than other state-of-the-art approaches under the premise of ensuring the identification accuracy. This work provides a new method for the online leakage monitoring of nuclear power pressure pipelines, and has important supporting significance for the online leakage monitoring of other large and complex equipment.

Improved Pedestrian Vehicle Detection for Small Objects Based on Attention Mechanism

Abstract This study aims to solve the low detection accuracy and susceptibility to false detection and omission in pedestrian and vehicle detection by proposing an improved YOLOv5s algorithm. Firstly, a small target detection module is added to better acquire and determine the information of pedestrians from long-range vehicles. Secondly, the multi-scale channel attention CBAM attention module is added, and the dual attention mechanism is not only flexible and convenient, but also improves the computational efficiency. Finally, the MPDIoU loss function based on minimum point distance is introduced to replace the original GIoU loss function, and this change not only enhances the regression accuracy of the model. At the same time, the convergence speed of the model is accelerated. KITTI data set was used for experiments, and the experimental results showed that the average accuracy of the model trained by the improved YOLOv5s algorithm on the data set reached 84.9%, which was 3.7% higher than that of the original YOLOv5s algorithm. It is verified that the model is suitable for high accuracy of pedestrian and vehicle recognition in complex environments, and has high value for promotion.

Automatic Aircraft Identification with High Precision from SAR Images Considering Multiscale Problems and Channel Information Enhancement

The SAR system possesses the ability to carry out all-day and all-weather imaging, which is highly valuable in the application of aircraft identification. However, aircraft identification from SAR images still faces great challenges due to speckle noise interference, multiscale problems, and complex background interference. To solve these problems, an efficient bidirectional path multiscale fusion and attention network (EBMA-Net) is proposed in this paper. It employs bidirectional connectivity to fuse the features of aircraft with different scales to perform the accurate detection of aircraft even when the background is highly complex. In the presented EBMA-Net, a module called efficient multiscale channel attention fusion (EMCA) and three parallel squeeze efficient channel attention (SECA) modules are proposed. In the EMCA module, the bidirectional paths are created by stacking upper and lower fusion modules, which effectively integrate shallow detailed features and deep semantic information. So, the detection performance of aircraft at different scales is improved. In the SECA module, the dependency relationship between feature channels is explicitly modeled, which can automatically learn the importance of different channels, prioritize key features, so as to improve the precision and robustness of aircraft identification. In the experiment, the public dataset of aircraft identification (i.e., SAR-AIRcraft-1.0, which is generated from the GF-3 satellite) from high-resolution SAR systems is used, and several other excellent target-detection networks are used for performance comparison, namely, YOLOv5s, YOLOv7, MGCAN, and EBPA2N. According to the results, the average aircraft detection accuracy of EBMA-Net is 91.31%, which is 4.5% higher than YOLOv7; and the false alarm rate is decreased by 5%. Its accuracy in the identification of aircraft can reach 95.6%, which is about 3.7% higher than YOLOv7. Therefore, the EBMA-Net obviously outperforms the other networks for aircraft detection and identification. The proposed EBMA-Net, which can capture the detailed information and better restrain the background interference, could also be used to perform the detection and identification of dense targets with different scales and background from SAR images.

Photometric stereo multi-information fusion unsupervised anomaly detection algorithm

Due to different materials, product surfaces are susceptible to light, shadow, reflection, and other factors. Coupled with the appearance of defects of various shapes and types, as well as dust, impurities, and other interfering influences, normal and abnormal samples are difficult to distinguish and a common problem in the field of defect detection. Given this, this paper proposes an end-to-end photometric stereo multi-information fusion unsupervised anomaly detection model. First, the photometric stereo feature generator is used to obtain normal, reflectance, depth, and other information to reconstruct the 3D topographic details of the object’s surface. Second, a multi-scale channel attention mechanism is constructed to fully use the feature associations of different layers of the backbone network, and the limited feature information is used to enhance the defect characterization ability. Finally, the original image is fused with normal and depth features to find the feature variability between defects and defects, as well as between defects and background. The feature differences between the source and clone networks are utilized to achieve multi-scale detection and improve detection accuracy. In this paper, the model performance is verified on the PSAD dataset. The experimental results show that the algorithm in this paper has higher detection accuracy compared with other algorithms. Among them, the multi-scale attention mechanism and multi-information fusion input improve the detection accuracy by 2.56% and 1.57%, respectively. In addition, the ablation experiments further validate the effectiveness of the detection algorithm in this paper.

INA-Net: An integrated noise-adaptive attention neural network for enhanced medical image segmentation

Skin cancer, a growing health concern, poses a significant threat. Automated segmentation of skin lesions is crucial for precise clinical diagnosis and treatment. However, challenges such as hair interference, vascular occlusion, and ambiguous lesion boundaries in dermoscopic images complicate accurate segmentation. Recent research has focused on capturing multiscale-enhanced global information across spatial and channel dimensions to improve the segmentation of ambiguous boundaries. Despite these advancements, performance remains suboptimal in the presence of complex disturbances. To address this, we propose INA-Net, an Integrated Noise-Adaptive Attention Neural Network for enhanced medical image segmentation. INA-Net proposes a lightweight noise-range attention module that integrates local and simulated noise features with global information to improve robustness against complex mixed noises. Additionally, our proposed Edge-aware Spatial Attention (ESA) and Multi-scale Channel Attention (MCA) enhance feature information and boundary perception. The proposed Dynamic Noise Encoding module and Fourier Wavelet Analysis (FW-Parser) further improve robustness by handling high-frequency noise components. INA-Net effectively manages hair interference, vascular occlusion, and ambiguous lesion boundaries, thereby providing critical support for accurate lesion identification. Evaluations on the ISIC2016, ISIC2018, and PH2 datasets demonstrate outstanding performance, with IOU scores of 87.53%, 84.74%, and 85.30%, respectively. These results surpass those of other models with comparable complexity and size.

A generative machine learning model for the 3D reconstruction of material microstructure and performance evaluation

The 3D reconstruction is generally defined as the process of capturing the shape and appearance of real objects. By reconstructing 3D digital model from a series of 2D slices, it brings considerable convenience to visualize internal structure and decipher structure-property relation of a material. Nowadays, the 3D reconstruction is becoming a cutting-edge technique in depicting the internal structure and evaluating the physical performance of targeted materials. Recent years, generative machine learning methods, such as generative adversarial networks (GAN), have achieved tremendous success in AI-generated physical content (AIGPC). However, lots of technical challenges remain, including oversimplified models, oversized dataset requirement and inefficient convergence. These difficulties are caused by the insufficient ability to capture detailed features and the inadequacy of the generated model quality. To this end, a novel generative model is developed, which combines the multiscale features of U-net and the synthesis ability of GANs. With the help of the multiscale channel aggregation module, the hierarchical feature aggregation module and the convolutional block attention module, our model is developed to capture the features of the material microstructure better. The loss function is refined by combining the image regularization loss with the Wasserstein distance loss. In addition, the anisotropy index is adopted to measure anisotropic degree of selected samples quantitively. The results demonstrate that the 3D structures generated by the proposed model retain high fidelity with ground truth samples. With remarkable performance, the proposed model not only overtakes traditional GAN, but will shed a brilliant light on AIGPC and physical 3D reconstruction.

A multi-scale channel attention network with federated learning for magnetic resonance image super-resolution

A multi-scale channel attention network with federated learning for magnetic resonance image super-resolution

DHAFormer: Dual-channel hybrid attention network with transformer for polyp segmentation.

The accurate early diagnosis of colorectal cancer significantly relies on the precise segmentation of polyps in medical images. Current convolution-based and transformer-based segmentation methods show promise but still struggle with the varied sizes and shapes of polyps and the often low contrast between polyps and their background. This research introduces an innovative approach to confronting the aforementioned challenges by proposing a Dual-Channel Hybrid Attention Network with Transformer (DHAFormer). Our proposed framework features a multi-scale channel fusion module, which excels at recognizing polyps across a spectrum of sizes and shapes. Additionally, the framework's dual-channel hybrid attention mechanism is innovatively conceived to reduce background interference and improve the foreground representation of polyp features by integrating local and global information. The DHAFormer demonstrates significant improvements in the task of polyp segmentation compared to currently established methodologies.

Mitigating Modality Discrepancies for RGB-T Semantic Segmentation.

Semantic segmentation models gain robustness against adverse illumination conditions by taking advantage of complementary information from visible and thermal infrared (RGB-T) images. Despite its importance, most existing RGB-T semantic segmentation models directly adopt primitive fusion strategies, such as elementwise summation, to integrate multimodal features. Such strategies, unfortunately, overlook the modality discrepancies caused by inconsistent unimodal features obtained by two independent feature extractors, thus hindering the exploitation of cross-modal complementary information within the multimodal data. For that, we propose a novel network for RGB-T semantic segmentation, i.e. MDRNet+, which is an improved version of our previous work ABMDRNet. The core of MDRNet+ is a brand new idea, termed the strategy of bridging-then-fusing, which mitigates modality discrepancies before cross-modal feature fusion. Concretely, an improved Modality Discrepancy Reduction (MDR+) subnetwork is designed, which first extracts unimodal features and reduces their modality discrepancies. Afterward, discriminative multimodal features for RGB-T semantic segmentation are adaptively selected and integrated via several channel-weighted fusion (CWF) modules. Furthermore, a multiscale spatial context (MSC) module and a multiscale channel context (MCC) module are presented to effectively capture the contextual information. Finally, we elaborately assemble a challenging RGB-T semantic segmentation dataset, i.e., RTSS, for urban scene understanding to mitigate the lack of well-annotated training data. Comprehensive experiments demonstrate that our proposed model surpasses other state-of-the-art models on the MFNet, PST900, and RTSS datasets remarkably.

(HTBNet)Arbitrary Shape Scene Text Detection with Binarization of Hyperbolic Tangent and Cross-Entropy.

The existing segmentation-based scene text detection methods mostly need complicated post-processing, and the post-processing operation is separated from the training process, which greatly reduces the detection performance. The previous method, DBNet, successfully simplified post-processing and integrated post-processing into a segmentation network. However, the training process of the model took a long time for 1200 epochs and the sensitivity to texts of various scales was lacking, leading to some text instances being missed. Considering the above two problems, we design the text detection Network with Binarization of Hyperbolic Tangent (HTBNet). First of all, we propose the Binarization of Hyperbolic Tangent (HTB), optimized along with which the segmentation network can expedite the initial convergent speed by reducing the number of epochs from 1200 to 600. Because features of different channels in the same scale feature map focus on the information of different regions in the image, to better represent the important features of all objects in the image, we devise the Multi-Scale Channel Attention (MSCA). Meanwhile, considering that multi-scale objects in the image cannot be simultaneously detected, we propose a novel module named Fused Module with Channel and Spatial (FMCS), which can fuse the multi-scale feature maps from channel and spatial dimensions. Finally, we adopt cross-entropy as the loss function, which measures the difference between predicted values and ground truths. The experimental results show that HTBNet, compared with lightweight models, has achieved competitive performance and speed on Total-Text (F-measure:86.0%, FPS:30) and MSRA-TD500 (F-measure:87.5%, FPS:30).

SPMUNet: Semantic segmentation of citrus surface defects driven by superpixel feature

The segmentation of surface defects in citrus fruit makes a significant contribution to quality assessment and post-harvest processing. However, semantic segmentation of citrus surface defects remains challenging owing to the complexity of defect features, unclear defect boundaries and semantic gaps in models. Therefore, a superpixel driven and multiscale channel cross-fusion transformer (MCCT)-based UNet (named SPMUNet) was proposed to segment citrus surface defects. The model first created a multiscale superpixel region feature (MSRF) branch using the simple linear iterative cluster (SLIC) algorithm to form the key features embedded in the downsampling layer. Then, the proposed Superpixel Loss optimized the loss function based on superpixel grid to provide error information for boundary segmentation. In addition, the MCCT module was used to improve the skip connection layers, and feature fusion was achieved by collaborative learning. Finally, the performance of the model was validated through ablation studies, feature map visualization, and comparison experiments. The results showed that the mPA, mIoU, and dice coefficients of SPMUNet for citrus surface segmentation were 89.1%, 83.8%, and 88.6%, respectively. Compared with the baseline, the proposed MSRF branch, MCCT, and Superpixel Loss improved mIoU by 2.6%, 1.3%, and 4.2%, respectively. And the effectiveness of the proposed methods was also validated by the feature map visualization. Furthermore, the proposed SPMUNet achieved optimal segmentation performance compared to other semantic segmentation models (UNet, ResUNet++, DeepLabv3+, SwinUNet, TransUNet). SPMUNet also excelled in the segmentation of small targets and boundaries, improving by 10.2% and 7.0%, respectively, over the baseline. The improvement in segmentation performance demonstrated the effectiveness of the selected superpixel features. The proposed citrus surface defect segmentation model achieved high-precision segmentation and serves as a technical reference for citrus sorting and quality assessment.

TMLNet: Triad Multitask Learning Network for multiobjective based change detection

Change detection is an essential computer vision task in remote sensing applications. It faces challenges of image registration errors, variation in image capturing conditions, clouds, etc. It is observed that an accurate change detection task requires effective feature learning, which needs a noiseless representation. Moreover, change detection improves with error-free image reconstruction as an auxiliary task, which needs joint feature representation by a feature extractor. In this work, we proposed a novel triad (which is a combination of input images and its difference) learning based multiresolution architecture TMLNet for effective change detection. We also proposed a multi-context local self-attention module to efficiently calculate long-range pixel relations with multiple contexts. An enhanced backbone module with top-down connections and multi-scale channel and spatial attention is utilized for change map generation. It provides less noisy features extracted through the backbone. Laplacian pyramid loss is used to preserve small details in feature reconstruction. A set of comprehensive experimentations reveals that the proposed scheme achieved the state-of-the-art result for the F1 Score, intersection over union, and overall accuracy values in seven benchmark datasets. Our model code is available at https://github.com/chouhan-avinash/TMLNet.

Environmentally adaptive fast object detection in UAV images

Detecting objects in aerial images poses a challenging task due to the presence of numerous small objects and complex environmental information. To address these problems, we propose an efficient detector specifically designed for aerial images, named EAF-YOLOv8, based on YOLOv8-S. In this paper, we introduce a novel backbone network called EAFNet, specifically designed for small object detection. EAFNet consists of the Rapidly Merging Receptive Fields Aggregation Module (RMRFAM) and Multi-Scale Channel Attention (MSCA). The RMRFAM utilizes dilated convolution (DConv) and partial convolution (PConv) to acquire richer receptive fields, capturing more extensive contextual information at higher levels while reducing redundancy in channel information, thereby accelerating inference speed. Furthermore, inspired by denoising tasks, we focus on the feature information surrounding the target background and propose MSCA. MSCA integrates channel attention with an embedded self-attention feature pyramid, extending the feature learning scope to the surrounding environment of the target, beyond the target itself. This approach utilizes enhanced background features to elicit a higher response for small targets, reducing false positives. Experimental results demonstrate that in UAVDT and VisDrone2019, the proposed EAF-YOLOv8 achieves mAP50 scores of 34.3% and 49.7%, respectively. Additionally, EAF-YOLOv8 exhibits high real-time inference speeds of 77.60 FPS and 55.56 FPS, showcasing competitive detection performance.

In-air handwriting system based on multi-scale channel attention network and monocular vision

In-air handwriting based on a monocular camera is an innovative and promising modality for human–computer interaction, offering a plethora of potential applications. However, existing in-air handwriting systems based on monocular camera suffer a significant challenge in determining the spatial coordinates of fingertips using two-dimensional images from a monocular camera. Since the size of the fingertip is very small and has very few discriminative features. To tackle this challenge, we propose a Multi-Scale Channel Attention Network (MSCAN). Through weighting multi-scale channels, the MSCAN facilitates the concentration of target detection models on high-resolution, small-scale channels, thus enhancing fingertip localization precision. We integrate the MSCAN with the YOLOv5s model to realize a novel in-air handwriting system based on monocular vision. We conducted comparative experiments on a self-constructed fingertip dataset and several publicly available small-target datasets. Experimental results show that the proposed method can effectively improve the accuracy of fingertip and small-target detection. The detection rate of fingertips reaches 98%, indicating that the proposed in-air handwriting system enables users to write freely and smoothly.

Enhancement of Mine Images through Reflectance Estimation of V Channel Using Retinex Theory

The dim lighting and excessive dust in underground mines often result in uneven illumination, blurriness, and loss of detail in surveillance images, which hinders subsequent intelligent image recognition. To address the limitations of the existing image enhancement algorithms in terms of generalization and accuracy, this paper proposes an unsupervised method for enhancing mine images in the hue–saturation–value (HSV) color space. Inspired by the HSV color space, the method first converts RGB images to the HSV space and integrates Retinex theory into the brightness (V channel). Additionally, a random perturbation technique is designed for the brightness. Within the same scene, a U-Net-based reflectance estimation network is constructed by enforcing consistency between the original reflectance and the perturbed reflectance, incorporating ResNeSt blocks and a multi-scale channel pixel attention module to improve accuracy. Finally, an enhanced image is obtained by recombining the original hue (H channel), brightness, and saturation (S channel), and converting back to the RGB space. Importantly, this image enhancement algorithm does not require any normally illuminated images during training. Extensive experiments demonstrated that the proposed method outperformed most existing unsupervised low-light image enhancement methods, qualitatively and quantitatively, achieving a competitive performance comparable to many supervised methods. Specifically, our method achieved the highest PSNR value of 22.18, indicating significant improvements compared to the other methods, and surpassing the second-best WCDM method by 10.3%. In terms of SSIM, our method also performed exceptionally well, achieving a value of 0.807, surpassing all other methods, and improving upon the second-place WCDM method by 19.5%. These results demonstrate that our proposed method significantly enhanced image quality and similarity, far exceeding the performance of the other algorithms.

ProLesA-Net: A multi-channel 3D architecture for prostate MRI lesion segmentation with multi-scale channel and spatial attentions

Prostate cancer diagnosis and treatment relies on precise MRI lesion segmentation, a challenge notably for small (<15mm) and intermediate (15-30mm) lesions. Our study introduces ProLesA-Net, a multi-channel 3D deep-learning architecture with multi-scale squeeze and excitation and attention gate mechanisms. Tested against six models across two datasets, ProLesA-Net significantly outperformed in key metrics: Dice score increased by 2.2%, and Hausdorff distance and average surface distance improved by 0.5mm, with recall and precision also undergoing enhancements. Specifically, for lesions under 15mm, our model showed a notable increase in five key metrics. In summary, ProLesA-Net consistently ranked at the top, demonstrating enhanced performance and stability. This advancement addresses crucial challenges in prostate lesion segmentation, enhancing clinical decision making and expediting treatment processes.

DMAF-NET: Deep Multi-Scale Attention Fusion Network for Hyperspectral Image Classification with Limited Samples

In recent years, deep learning methods have achieved remarkable success in hyperspectral image classification (HSIC), and the utilization of convolutional neural networks (CNNs) has proven to be highly effective. However, there are still several critical issues that need to be addressed in the HSIC task, such as the lack of labeled training samples, which constrains the classification accuracy and generalization ability of CNNs. To address this problem, a deep multi-scale attention fusion network (DMAF-NET) is proposed in this paper. This network is based on multi-scale features and fully exploits the deep features of samples from multiple levels and different perspectives with an aim to enhance HSIC results using limited samples. The innovation of this article is mainly reflected in three aspects: Firstly, a novel baseline network for multi-scale feature extraction is designed with a pyramid structure and densely connected 3D octave convolutional network enabling the extraction of deep-level information from features at different granularities. Secondly, a multi-scale spatial–spectral attention module and a pyramidal multi-scale channel attention module are designed, respectively. This allows modeling of the comprehensive dependencies of coordinates and directions, local and global, in four dimensions. Finally, a multi-attention fusion module is designed to effectively combine feature mappings extracted from multiple branches. Extensive experiments on four popular datasets demonstrate that the proposed method can achieve high classification accuracy even with fewer labeled samples.

Semantic Coarse-to-Fine Granularity Learning for Two-Stage Few-Shot Anomaly Detection

Anomaly detection is critical in industrial inspection, where identifying defects significantly impacts product quality and safety. Existing models, primarily based on convolutional neural networks (CNNs), struggle with noise sensitivity and insufficient resolution for fine-grained feature discrimination. To address these issues, we propose a two-stage few-shot anomaly detection network that enhances semantic feature granularity and generalization. The network includes a coarse-grained anomaly detection module, a multi-scale channel attention module, and a fine-grained detection module. The coarse-grained module identifies abnormal regions, serving as the initial filter. The multi-scale channel attention module focuses on anomalous features, enhancing sensitivity to fine-grained characteristics. This step overcomes limitations in discerning subtle yet critical anomalies. The fine-grained detection module refines feature maps, enhancing generalization. Experimental results on the MVTec dataset show an image-level Area under the region of convergence (AUROC) of 92.3% and a pixel-level AUROC of 95.3%, a 1% to 2% improvement over leading FSAD methods.