Multi-scale Feature Selection Research Articles

Hyperspectral Image Classification Algorithm for Forest Analysis Based on a Group-Sensitive Selective Perceptual Transformer

Substantial advancements have been achieved in hyperspectral image (HSI) classification through contemporary deep learning techniques. Nevertheless, the incorporation of an excessive number of irrelevant tokens in large-scale remote sensing data results in inefficient long-range modeling. To overcome this hurdle, this study introduces the Group-Sensitive Selective Perception Transformer (GSAT) framework, which builds upon the Vision Transformer (ViT) to enhance HSI classification outcomes. The innovation of the GSAT architecture is primarily evident in several key aspects. Firstly, the GSAT incorporates a Group-Sensitive Pixel Group Mapping (PGM) module, which organizes pixels into distinct groups. This allows the global self-attention mechanism to function within these groupings, effectively capturing local interdependencies within spectral channels. This grouping tactic not only boosts the model’s spatial awareness but also lessens computational complexity, enhancing overall efficiency. Secondly, the GSAT addresses the detrimental effects of superfluous tokens on model efficacy by introducing the Sensitivity Selection Framework (SSF) module. This module selectively identifies the most pertinent tokens for classification purposes, thereby minimizing distractions from extraneous information and bolstering the model’s representational strength. Furthermore, the SSF refines local representation through multi-scale feature selection, enabling the model to more effectively encapsulate feature data across various scales. Additionally, the GSAT architecture adeptly represents both global and local features of HSI data by merging global self-attention with local feature extraction. This integration strategy not only elevates classification precision but also enhances the model’s versatility in navigating complex scenes, particularly in urban mapping scenarios where it significantly outclasses previous deep learning methods. The advent of the GSAT architecture not only rectifies the inefficiencies of traditional deep learning approaches in processing extensive remote sensing imagery but also markededly enhances the performance of HSI classification tasks through the deployment of group-sensitive and selective perception mechanisms. It presents a novel viewpoint within the domain of hyperspectral image classification and is poised to propel further advancements in the field. Empirical testing on six standard HSI datasets confirms the superior performance of the proposed GSAT method in HSI classification, especially within urban mapping contexts, where it exceeds the capabilities of prior deep learning techniques. In essence, the GSAT architecture markedly refines HSI classification by pioneering group-sensitive pixel group mapping and selective perception mechanisms, heralding a significant breakthrough in hyperspectral image processing.

Apple recognition in complex environments based on FC-DETR

This study aims to address the challenges of apple recognition in complex environments by proposing a solution based on the FC-DETR model. Apple cultivation is an important part of agriculture in Xinjiang, but the increasing shortage of labor has driven the demand for automated harvesting technologies. Therefore, research and development of apple recognition technology in complex environments have become crucial. While CNN architecture models can accurately identify apples in conventional settings, their inherent receptive field limitations prevent them from fully capturing global features in noisy and complex environments, making it difficult to achieve the accuracy and robustness required for practical applications. To solve this problem, this study proposes a real-time object detection model, FC-DETR. This model incorporates the innovative FEMA-BasicBlock residual module, the CAFM cross-scale adaptive feature fusion module, and the novel Inner-WIoU loss function, thereby enhancing feature processing, multi-scale feature selection and integration, and detection accuracy. Experimental results show that the FC-DETR model achieves an apple recognition accuracy of 87 % and a recall rate of 82 % in complex backgrounds while maintaining a lightweight design. This study not only makes significant advances in automated apple harvesting technology but also contributes to improving the efficiency and sustainability of the apple industry in Xinjiang and globally.

Convolutional Neural Network-Based Multiscale Feature Selection and Evaluation in Image Segmentation

Convolutional Neural Network-Based Multiscale Feature Selection and Evaluation in Image Segmentation

Modeling the distribution of headwater streams using topoclimatic indices, remote sensing and machine learning.

Headwater streams (HWS) are ecologically important components of montane ecosystems. However, they are difficult to map and may not be accurately represented in existing spatial datasets. We used topographically resolved climatic water balance data and satellite indices retrieved from Google Earth Engine to model the occurrence (presence or absence) of HWS across Northwest Montana. A multi-scale feature selection (MSFS) procedure and boosted regression tree models/machine learning algorithms were used to identify variables associated with HWS occurrence. In final model evaluation, models that included climatic water balance deficit were more accurate (83.5% ranging from 82.9% to 83.7%) than using only terrain indices (81.1% ranging from 80.7% to 81.4%) and improved upon estimates of stream extent represented by the National Hydrography Dataset Plus High Resolution (NHDPlus HR) (82.7% ranging from 82.5% to 83.1%). Including topoclimate captured the varying effect of upslope accumulated area across a strong moisture gradient. Multi-scale cross-validation, coupled with a MSFS algorithm allowed us to find a parsimonious model that was not immediately evident using standard cross-validation procedures. More accurate spatial model predictions of HWS have potential for immediate application in land and water resource management, where significant field time can be spent identifying potential stream impacts prior to contracting and planning.

Multi-scale feature selection network for lightweight image super-resolution

Recently, many super-resolution (SR) methods based on convolutional neural networks (CNNs) have achieved superior performance by utilizing deep and heavy models, which may not be suitable for real-world low-budget devices. To address this issue, we propose a novel lightweight SR network called a multi-scale feature selection network (MFSN). As the basic building block of MFSN, the multi-scale feature selection block (MFSB) is presented to extract the rich multi-scale features from a coarse-to-fine receptive field level. For a better representation ability, a wide-activated residual unit is adopted in each branch of MFSB except the last one. In MFSB, the scale selection module (SSM) is designed to effectively fuse the features from two adjacent branches by adjusting receptive field sizes adaptively. Further, a comprehensive channel attention mechanism (CCAM) is integrated into SSM to learn the dynamic selection weight by considering the local and global inter-channel dependencies. Extensive experimental results illustrate that the proposed MFSN is superior to other lightweight methods.

Attention-based dual-path feature fusion network for automatic skin lesion segmentation

Automatic segmentation of skin lesions is a critical step in Computer Aided Diagnosis (CAD) of melanoma. However, due to the blurring of the lesion boundary, uneven color distribution, and low image contrast, resulting in poor segmentation result. Aiming at the problem of difficult segmentation of skin lesions, this paper proposes an Attention-based Dual-path Feature Fusion Network (ADFFNet) for automatic skin lesion segmentation. Firstly, in the spatial path, a Boundary Refinement (BR) module is designed for the output of low-level features to filter out irrelevant background information and retain more boundary details of the lesion area. Secondly, in the context path, a Multi-scale Feature Selection (MFS) module is constructed for high-level feature output to capture multi-scale context information and use the attention mechanism to filter out redundant semantic information. Finally, we design a Dual-path Feature Fusion (DFF) module, which uses high-level global attention information to guide the step-by-step fusion of high-level semantic features and low-level detail features, which is beneficial to restore image detail information and further improve the pixel-level segmentation accuracy of skin lesion. In the experiment, the ISIC 2018 and PH2 datasets are employed to evaluate the effectiveness of the proposed method. It achieves a performance of 0.890/ 0.925 and 0.933 /0.954 on the F1-score and SE index, respectively. Comparative analysis with state-of-the-art segmentation methods reveals that the ADFFNet algorithm exhibits superior segmentation performance.

MFSPest: A multi-scale feature selection network for light-trapped agricultural pest detection

Accurate and fast pest detection is crucial for ensuring high crop yield and quality in modern agriculture. However, there are significant challenges in using deep learning for pest detection, such as the small proportion of pest individuals in the image area, complex backgrounds in light-trapped pest images, and an unbalanced distribution of pest species. To address these problems, we propose MFSPest, a multi-scale feature selection network for detecting agricultural pests in trapping scenes. We design a novel selective kernel spatial pyramid pooling structure (SKSPP) in the feature extraction stage to enhance the network’s feature extraction ability for key regions and reduce its focus on irrelevant background information. Furthermore, we present the equalized loss to increase the loss weights of rare categories and improve the distribution imbalance among pest categories. Finally, we build LAPD, a light-trapping agricultural pest dataset containing nine pest categories. We conducted experiments on this dataset and demonstrated that our proposed method achieves state-of-the-art performance, with Accuracy, Recall, and mean Average Precision (mAP) of 89.9%, 92.8%, and 93.6%, respectively. Our method satisfies the requirements of pest detection applications in practical scenarios and has practical value and economic benefits for use in agricultural pest trapping and management.

Quantifying scattering characteristics of mangrove species from Optuna-based optimal machine learning classification using multi-scale feature selection and SAR image time series

Mangroves play a significant role in carbon sequestration and storage. Mapping mangrove species and monitoring their conditions have been a crucial issue for achieving sustainable development goals. Currently combing multidimensional optical and SAR images with machine learning have become an important approach for mangrove species classification, but there are still some challenges in feature selection and hyperparameter optimizations. In this study, we proposed a novel classification framework by combing multi-scale variable selection algorithm (MUVR) with state-of-the-art machine learning hyperparameter optimization method (Optuna) for mapping mangrove species in the Beilun Estuary and Maowei Sea nature reserves using optical and dual-polarization SAR images, and further quantified the scattering characteristics of mangrove species using SAR image time series. We found that: (1) The MUVR algorithm could determine the optimal scale features for different scenarios and mangrove species, and improve the classification performance of machine learning with an overall accuracy (OA) improvement of 12.85%; (2) The Optuna-based optimal CatBoost outperforms LightGBM and NGBoost algorithms in mapping mangrove species, which achieved the highest OA (93.18%). This study demonstrated that LightGBM was suitable for identifying Aegiceras corniculatum, while the CatBoost algorithm was suitable for discriminating Avicennia marina, Bruguiera gymnorrhiza, Cyperus malaccensis, Kandelia candel and Sonneratia apetala; (3) SAR images and its derivatives improved identification ability of mangrove species, and collaboration of multispectral images and SAR-derived features produced the better classification; (4) From 2018 to 2020, the backscattering coefficients of mangrove species in VV and VH polarization focused on 0.053–0.327 and 0.015–0.062, respectively. The coherence coefficients of mangroves displayed a seasonal change trend with the large variations in summer and small variations in winter. The range of Entropy and Alpha of mangrove species was from 0.65 to 0.88 and 17–33, which indicated that the main scattering mechanism of mangroves was moderate random surface scattering.

CAMS-Net: An attention-guided feature selection network for rib segmentation in chest X-rays

Segmentation of clavicles and ribs in chest X-rays is significant for diagnosing lung diseases. However, it is challenging to segment ribs because of the low contrast on chest X-rays. Moreover, most existing methods fail to segment ribs in the area of abnormal gray value caused by overlapping anatomical structures or lesions. A novel algorithm based on attention-guided feature selection named CAMS-Net is presented in this paper, aiming to improve the accuracy of ribs segmentation in low contrast areas and abnormal gray value areas. The collaborative attention skip connection module (CAS) introduces the decoder features and attention-guided feature selection into the traditional skip connection, which highlights the feature representation of ribs. The attention-guided multi-scale feature selection module (AMFS) increases receptive field size to connect the abnormal rib gray value region with the normal gray value region. To reduce the influence of background pixels, the AMFS selects important features through attention and uses multi-scale information to jointly decide the final class of the pixel. The paper conducted extensive experiments to evaluate CAMS-Net. Compared with state-of-the-art methods, the average values of Precision and Jaccard are increased by 0.89% and 1.23%, respectively. The Recall and Jaccard values of the anterior rib are increased by 2.64% and 2.52%, respectively. Qualitative analysis shows that CAMS-Net can improve the segmentation accuracy of low-contrast areas and maintain the segmentation integrity in areas with abnormal gray values. The robustness and generalization of CAMS-Net are verified through external tests on VinDr-RibCXR, JSRT, Shenzhen, and NIH datasets. Besides, CAS and AMFS modules can be flexibly inserted into other backbone networks.

TANet: Triple Attention Network for medical image segmentation

In recent years, deep learning-based methods have achieved remarkable progress in medical image processing, like polyp segmentation in colonoscopy images and skin lesion segmentation in dermoscopy images. However, the current state-of-the-art medical segmentation methods still suffer from the problem of low accuracy in segmenting the small-scale and variable-scale objects. To solve this problem, we propose Triple Attention Network (TANet). In TANet, a novel Triple Attention Module (TAM) is presented. TAM has two sub-modules: Multi-scale Feature Selection Module (MFSM) and Contextual Feature Extraction Module (CFEM). MFSM is used to extract more adaptable multi-scale features for capturing variable-scale objects, while CFEM is for capturing small-scale objects by extracting contextual features. TAM aims to combine MFSM and CFEM to finally enhance the segmentation performance of the medical images with the small-scale and variable-scale lesions. Extensive experiments are conducted on five polyp datasets and one skin lesion dataset. Results show that the proposed models outperform the previous state-of-the-art models on most evaluation metrics and improve the Dice score by up to 7.1%. All results consistently confirm the effectiveness of the proposed TANet and show that the TANet achieves state-of-the-art performance on the above datasets.

MDF-Net: A Multi-Scale Dynamic Fusion Network for Breast Tumor Segmentation of Ultrasound Images.

Breast tumor segmentation of ultrasound images provides valuable information of tumors for early detection and diagnosis. Accurate segmentation is challenging due to low image contrast between areas of interest; speckle noises, and large inter-subject variations in tumor shape and size. This paper proposes a novel Multi-scale Dynamic Fusion Network (MDF-Net) for breast ultrasound tumor segmentation. It employs a two-stage end-to-end architecture with a trunk sub-network for multiscale feature selection and a structurally optimized refinement sub-network for mitigating impairments such as noise and inter-subject variation via better feature exploration and fusion. The trunk network is extended from UNet++ with a simplified skip pathway structure to connect the features between adjacent scales. Moreover, deep supervision at all scales, instead of at the finest scale in UNet++, is proposed to extract more discriminative features and mitigate errors from speckle noise via a hybrid loss function. Unlike previous works, the first stage is linked to a loss function of the second stage so that both the preliminary segmentations and refinement subnetworks can be refined together at training. The refinement sub-network utilizes a structurally optimized MDF mechanism to integrate preliminary segmentation information (capturing general tumor shape and size) at coarse scales and explores inter-subject variation information at finer scales. Experimental results from two public datasets show that the proposed method achieves better Dice and other scores over state-of-the-art methods. Qualitative analysis also indicates that our proposed network is more robust to tumor size/shapes, speckle noise and heavy posterior shadows along tumor boundaries. An optional post-processing step is also proposed to facilitate users in mitigating segmentation artifacts. The efficiency of the proposed network is also illustrated on the "Electron Microscopy neural structures segmentation dataset". It outperforms a state-of-the-art algorithm based on UNet-2022 with simpler settings. This indicates the advantages of our MDF-Nets in other challenging image segmentation tasks with small to medium data sizes.

Ultra-short-term probabilistic wind power forecasting with spatial-temporal multi-scale features and K-FSDW based weight

As a potential cleaner energy technology, wind power is a pollution-free and inexhaustible energy, which make a significant contribution to the global energy transformation. Most studies have focused on the accurate forecasting to help the management of the wind power grid-tied. Considering the need for the quantitative modeling of the endogenous random fluctuations and uncertainties involved, a novel ultra-short-term probabilistic wind power forecasting with spatial–temporal multi-scale features and K-FSDW based weight is proposed, which includes data decomposition, multi-scale feature selection, individual model training, dynamic weighting using K-FSDW, error correction modeling by RF and probability density forecasting by KDE, to create uncertainty quantification for power grid dispatching and operation. Firstly, the normalized target wind speed is decomposed into multiple subsequences through VMD, the subsequence and adjacent spatial wind speed series are reconstructed into spatiotemporal candidate features, and spatial–temporal multi-scale feature selection is carried out. Secondly, different quantile regression models are used to predict each subsequence, and a quantile dynamic sparse weighted combination algorithm based on K-forward nearest neighbor is proposed to combine the prediction results of each model, and then reorganize the prediction results of subsequences. Finally, the RF model is used for error correction, and the probability density function is obtained by kernel density estimation. In the experimental comparison and analysis, taking the actual data of an offshore wind farm in Penglai District, Shandong Province, China as an example, the feasibility and effectiveness of the model are verified.

ADL-CDF: A Deep Learning Framework for COVID-19 Detection from CT Scans Towards an Automated Clinical Decision Support System.

The emergence of deep learning has paved to solve many problems in the real world. COVID-19 pandemic, since the late 2019, has been affecting lives of people across the globe. Chest CT scan images are used to detect it and know its severity in patients. The problem with many existing solutions in COVID-19 detection using CT scan images is that inability to detect the infection when it is in initial stages. As the infection can exist on varied scales, there is need for more comprehensive approach that can ascertain the disease at all scales. Towards this end, we proposed a deep learning-based framework known as Automated Deep Learning-based COVID-19 Detection Framework (ADL-CDF). It does not need a human medical expert in diagnosis as it is capable of detecting automatically. The framework is assisted by two algorithms that involve image processing and deep learning. The first algorithm known as Region of Interest (ROI)-based Image Filtering (ROI-IF) which analyses given input CT scan images of a patient and discards the ones where ROI is missing. This algorithm minimizes time taken for processing besides reducing false positive rate. The second algorithm is known as Multi-Scale Feature Selection algorithm that fits into the deep learning framework’s pipeline to leverage detection performance of the ADL-CDF. The proposed framework is evaluated against ResNet50V2 and Xception. Our empirical study revealed that our model outperforms the state of the art.

A multi-scale feature selection approach for predicting benthic assemblages

Seafloor habitat maps are an important management tool used to delineate distinct regions of the seabed based on their biophysical properties. Spatially continuous bathymetry and backscatter-derived terrain features are commonly used as proxies for environmental conditions and processes that affect species distributions. Multi-scale approaches are increasingly applied to assess the relevant scales at which species co-occur. As the optimal scale(s) may be unknown, features can be calculated at multiple successive scales, yet this results in numerous highly correlated features that may negatively impact model interpretability. To address this increased dimensionality, feature selection approaches can be used to identify the most relevant features. Here, filter and wrapper approaches are assessed to select features from a highly dimensional multi-scale dataset. Terrain features describing the seabed were calculated across ten scales at two coastal sites in Placentia Bay, Newfoundland, Canada. Five species assemblages were identified using ground-truth underwater video sampling. Features predicting the presence of assemblages were assessed using the two selection methods, and the set of chosen features was modelled using three machine learning algorithms: extreme gradient boosting (XGB), random forest (RF), and support vector machines (SVM). The XGB model with features selected by scale-factor from the Boruta wrapper algorithm had the highest accuracy according to cross-validation- (61.67%, kappa 0.49). Bathymetry and terrain attributes were the most important predictors of assemblage occurrence across various analysis scales encompassing both broader and fine-scale variability of the seabed. The proposed feature reduction and selection approach improved the overall accuracy of predictions, and the resulting biological complexity captured in our habitat maps established baseline data for an ecologically significant coastal region.

SLMS-SSD: Improving the balance of semantic and spatial information in object detection

With the development of deep learning technology, the research on convolutional neural network-based object detection is becoming more and more mature. However, most methods are unsatisfactory in dealing with the issue of semantic and spatial information imbalance. In this article, we extend the single-shot multibox detector SSD and propose a self-learning multi-scale object detection network by balancing the semantic information and spatial information, named SLMS-SSD. We first construct a shallow feature enhancement module to enhance the representation of small objects by extracting richer context information. Second, in terms of feature connectivity, we design a multi-scale feature selection module for intermediate layer features with a combination of top-down and direct up-samplings. Finally, in terms of feature strength, we design a self-learning feature fusion module for measuring the feature importance. We validate our model on the PASCAL VOC and MS COCO datasets, and the results demonstrate that it can effectively improve the accuracy of object detection, especially the accuracy of small object detection.

Target based image fusion using multi-scale feature selection in three regions

Image captured by low dynamic range (LDR) camera fails to capture entire exposure level of scene, and instead only covers certain range of exposures. In order to cover entire exposure level in single image, bracketed exposure LDR images are combined. The range of exposures in different images results in information loss in certain regions. These regions need to be addressed and based on this motive a novel methodology of layer based fusion is proposed to generate high dynamic range image. High and low-frequency layers are formed by dividing each image based on pixel intensity variations. The regions are identified based on information loss section created in differently exposed images. High-frequency layers are combined using region based fusion with Dense SIFT which is used as activity level testing measure. Low-frequency layers are combined using weighted sum. Finally combined high and low-frequency layers are merged together on pixel to pixel basis to synthesize fused image. Objective analysis is performed to compare the quality of proposed method with state-of-the-art. The measures indicate superiority of the proposed method.

A Multi-Scale Feature Selection Method for Steganalytic Feature GFR

The Rich Model of the Gabor filter (referred to as the GFR steganalytic feature) can detect JPEG-adaptive steganography objects. However, feature dimensionality that is too high will lead to too much computation and will correspondingly reduce the detection efficiency. To reduce the dimensionality and the operating time of GFR steganalytic features and to improve the stego image detection accuracy, this paper proposes a multi-scale feature selection method for steganalytic feature GFR. First, we use the SNR criterion to measure the uselessness of each feature component and to provide a basis for the removal of useless steganalytic feature components. Second, we improve the Relief algorithm to measure the importance of feature components in detecting stego images, which provides a basis for the selection of important feature components. Then, we set the threshold value for deleting the useless feature components, and we select the important feature components as the final feature. Finally, we conduct experiments on feature selection for GFR with high-dimensional steganalytic features, and we compared the proposed method with the Fisher-based method, the PCA-based method, the SSFC method, and the Steganalysis-α method. The results show that the method proposed in this paper is effective and fast.

OBJECT-ORIENTED CHANGE DETECTION BASED ON MULTI-SCALE APPROACH

The change detection of remote sensing images means analysing the change information quantitatively and recognizing the change types of the surface coverage data in different time phases. With the appearance of high resolution remote sensing image, object-oriented change detection method arises at this historic moment. In this paper, we research multi-scale approach for high resolution images, which includes multi-scale segmentation, multi-scale feature selection and multi-scale classification. Experimental results show that this method has a stronger advantage than the traditional single-scale method of high resolution remote sensing image change detection.

OBJECT-ORIENTED CHANGE DETECTION BASED ON MULTI-SCALE APPROACH

Abstract. The change detection of remote sensing images means analysing the change information quantitatively and recognizing the change types of the surface coverage data in different time phases. With the appearance of high resolution remote sensing image, object-oriented change detection method arises at this historic moment. In this paper, we research multi-scale approach for high resolution images, which includes multi-scale segmentation, multi-scale feature selection and multi-scale classification. Experimental results show that this method has a stronger advantage than the traditional single-scale method of high resolution remote sensing image change detection.