Low-level Feature Extraction Research Articles

A Blind Image Quality Index for Synthetic and Authentic Distortions with Hierarchical Feature Fusion

Blind Image Quality Assessment (BIQA) for synthetic and authentic distortions has attracted much attention in the community, and it is still a great challenge. The existing quality metrics are mildly consistent with subjective perception. Traditional handcrafted quality metrics can easily and directly extract low-level features, which mainly account for the outline, edge, color, texture, and shape features, while ignoring the important deep semantics of the distorted image. In the field of popular deep learning, multilevel features can be acquired easily. However, most of them either use only high-level features, ignoring the shallow features, or they simply combine features at different levels, resulting in limited prediction performance. Motivated by these, this paper presents a novel BIQA for synthetic and authentic distortions with hierarchical feature fusion in a flexible vision-Transformer framework. First, multiscale features are extracted from a strong vision-Transformer backbone. Second, an effective hierarchical feature fusion module is proposed to incorporate the features at different levels progressively. To eliminate redundant information, a simple but effective attention mechanism is employed after each fusion. Third, inspired by the human visual system, local and global features from the fusion features are extracted to represent different granularity distortions. Finally, these local and global features are mapped to the final quality score. Extensive experiments on three authentic image databases and two synthetic image datasets show that the proposed method is superior to the state-of-the-art quality metrics for both single-database testing and cross-database testing.

Spatial Anomaly Detection in Hyperspectral Imaging Using Optical Neural Networks

Hyperspectral imaging is a widely used technology, yet hard to implement in real-time anomaly detection due to its extensive data flow volume. An auto-encoder structured hybrid optical-electrical neural network method is proposed in this work that realizes feature exaction and anomaly detection during the hyperspectral data acquisition process to address such issues. In the proposed method, a digital micromirror device functions as the core optical processor to extract low-level features from the hyperspectral data flow. Weight binarization and a conditional sub-network are utilized to suit optical computation. Pretraining by artificial data is implemented to ease the training data burden. Case studies on defect detection and foreign object detection have demonstrated that the proposed method can significantly reduce the sampling time by orders of magnitude without loss of detection accuracy.

A Survey of Farmland Boundary Extraction Technology Based on Remote Sensing Images

Farmland boundary information plays a key role in agricultural remote sensing, and it is of importance to modern agriculture. We collected the relevant research in this field at home and abroad in this review, and we systematically assessed the farmland boundary extraction process, detection algorithms, and influencing factors. In this paper, we first discuss the five parts of the assessment: (1) image acquisition; (2) preprocessing; (3) detection algorithms; (4) postprocessing; (5) the evaluation of the boundary information extraction process. Second, we discuss recognition algorithms. Third, we discuss various detection algorithms. The detection algorithms can be divided into four types: (1) low-level feature extraction algorithms, which only consider the boundary features; (2) high-level feature extraction algorithms, which consider boundary information and other image information simultaneously; (3) visual hierarchy extraction algorithms, which simulate biological vision systems; (4) boundary object extraction algorithms, which recognize boundary object extraction ideas. We can subdivide each type of algorithm into several algorithm subclasses. Fourth, we discuss the technical factors and natural factors that affect boundary extraction. Finally, we summarize the development history of this field, and we analyze the problems that exist, such as the lack of algorithms that can be adapted to higher-resolution images, the lack of algorithms with good practical ability, and the lack of a unified and effective evaluation index system.

An Optimized Deep Learning Model with Feature Fusion for Brain Tumor Detection

The automatic detection of brain tumor from large volumes of MRI images using deep learning is a issue that necessitates substantial computing resources. So,in this study, a brain tumor detection framework using feature fusion from optimized shallow and deep learning models is proposed that efficiently detects different types of tumors. The human brain is a 3D object and the intensity of abnormal tissue varies as per location. As a result, incorporating surrounding tissue into tumor region can help to discriminate between the type of tumor and its severity. To extract deep characteristics from tumor region and adjacent tissues, deep models such as Inception-V3 is employed using transfer learning. Deep features are especially important in tumour detection, however as the network size grows, certain low-level insights about tumor are lost. As a result, a novel optimized shallow model is designed to extract low-level features. To overcome this limitation of information loss, deep and shallow features are fused. Our extensive simulation and experiment done on a publicly available benchmark dataset shows that an optimized hybrid deep learning model with ROI expansion improves tumor detection accuracy by 9\%. These findings support the theory that the tissues adjacent to the tumor contain unique information and feature fusion compensates for information loss.

MRI brain tumor detection and classification using parallel deep convolutional neural networks

Convolutional neural network (CNN) is widely used to classify brain tumors with high accuracy. Since CNN collects features randomly without knowing the local and global features and causes overfitting problems, this research proposes a novel parallel deep convolutional neural network (PDCNN) topology to extract both global and local features from the two parallel stages and deal with the over-fitting problem by utilizing dropout regularizer alongside batch normalization. To begin, input images are resized and grayscale transformation is conducted, which helps to reduce complexity. After that, data augmentation has been used to maximize the number of datasets. The benefits of parallel pathways are provided by combining two simultaneous deep convolutional neural networks with two different window sizes, allowing this model to learn local and global information. Three forms of MRI datasets are used to determine the effectiveness of the proposed method. The binary tumor identification dataset-I, Figshare dataset-II, and Multiclass Kaggle dataset-III provide accuracy of 97.33%, 97.60%, and 98.12%, respectively. The proposed structure is not only accurate but also efficient, as the proposed method extracts both low-level and high-level features, improving results compared to state-of-the-art techniques.

A hyperspectral image reconstruction algorithm based on RGB image using multi-scale atrous residual convolution network

Hyperspectral images are a valuable tool for remotely sensing important characteristics of a variety of landscapes, including water quality and the status of marine disasters. However, hyperspectral data are rare or expensive to obtain, which has spurred interest in low-cost, fast methods for reconstructing hyperspectral data from much more common RGB images. We designed a novel algorithm to achieve this goal using multi-scale atrous convolution residual network (MACRN). The algorithm includes three parts: low-level feature extraction, high-level feature extraction, and feature transformation. The high-level feature extraction module is composed of cascading multi-scale atrous convolution residual blocks (ACRB). It stacks multiple modules to form a depth network for extracting high-level features from the RGB image used as an input. The algorithm uses jump connection for residual learning, and the final high-level feature combines the output of the low-level feature extraction module and the output of the cascaded atrous convolution residual block element by element, so as to prevent gradient dispersion and gradient explosion in the deep network. Without adding too many parameters, the model can extract multi-scale features under different receptive fields, make better use of the spatial information in RGB images, and enrich the contextual information. As a proof of concept, we ran an experiment using the algorithm to reconstruct hyperspectral Sentinel-2 satellite data from the northern coast of Australia. The algorithm achieves hyperspectral spectral reconstruction in 443nm-2190nm band with less computational cost, and the results are stable. On the Realworld dataset, the reconstruction error MARE index is less than 0.0645, and the reconstruction time is less than 9.24S. Therefore, in the near infrared band, MACRN reconstruction accuracy is significantly better than other spectral reconstruction algorithms. MACRN hyperspectral reconstruction algorithm has the characteristics of low reconstruction cost and high reconstruction accuracy, and its advantages in ocean spectral reconstruction are more obvious.

TRAL: A Tag-Aware Recommendation Algorithm Based on Attention Learning

A social tagging system improves recommendation performance by introducing tags as auxiliary information. These tags are text descriptions of target items provided by individual users, which can be arbitrary words or phrases, so they can provide more abundant information about user interests and item characteristics. However, there are many problems to be solved in tag information, such as data sparsity, ambiguity, and redundancy. In addition, it is difficult to capture multi-aspect user interests and item characteristics from these tags, which is essential to the recommendation performance. In the view of these situations, we propose a tag-aware recommendation model based on attention learning, which can capture diverse tag-based potential features for users and items. The proposed model adopts the embedding method to produce dense tag-based feature vectors for each user and each item. To compress these vectors into a fixed-length feature vector, we construct an attention pooling layer that can automatically allocate different weights to different features according to their importance. We concatenate the feature vectors of users and items as the input of a multi-layer fully connected network to learn non-linear high-level interaction features. In addition, a generalized linear model is also conducted to extract low-level interaction features. By integrating these features of different types, the proposed model can provide more accurate recommendations. We establish extensive experiments on two real-world datasets to validate the effect of the proposed model. Comparable results show that our model perform better than several state-of-the-art tag-aware recommendation methods in terms of HR and NDCG metrics. Further ablation studies also demonstrate the effectiveness of attention learning.

Combining convolutional neural networks and self-attention for fundus diseases identification

Early detection of lesions is of great significance for treating fundus diseases. Fundus photography is an effective and convenient screening technique by which common fundus diseases can be detected. In this study, we use color fundus images to distinguish among multiple fundus diseases. Existing research on fundus disease classification has achieved some success through deep learning techniques, but there is still much room for improvement in model evaluation metrics using only deep convolutional neural network (CNN) architectures with limited global modeling ability; the simultaneous diagnosis of multiple fundus diseases still faces great challenges. Therefore, given that the self-attention (SA) model with a global receptive field may have robust global-level feature modeling ability, we propose a multistage fundus image classification model MBSaNet which combines CNN and SA mechanism. The convolution block extracts the local information of the fundus image, and the SA module further captures the complex relationships between different spatial positions, thereby directly detecting one or more fundus diseases in retinal fundus image. In the initial stage of feature extraction, we propose a multiscale feature fusion stem, which uses convolutional kernels of different scales to extract low-level features of the input image and fuse them to improve recognition accuracy. The training and testing were performed based on the ODIR-5k dataset. The experimental results show that MBSaNet achieves state-of-the-art performance with fewer parameters. The wide range of diseases and different fundus image collection conditions confirmed the applicability of MBSaNet.

Statistical Texture Learning Method for Monitoring Abandoned Suburban Cropland Based on High-Resolution Remote Sensing and Deep Learning

Cropland abandonment is crucial in agricultural management and has a profound impact on crop yield and food security. In recent years, many cropland abandonment identification methods based on remote sensing observation data have been proposed, but most of these methods are based on coarse-resolution images and use traditional machine learning methods for simple identification. To this end, we perform abandonment recognition on high-resolution remote sensing images. According to the texture features of the abandoned land, we combine the method of statistical texture learning and propose a new deep learning framework called PSPNet-STL. The model integrates high-level semantic feature extraction and deep mining of low-level texture features to identify cropland abandonment. First, we labeled the abandoned cropland area and built the HRAC dataset, a high-resolution cropland abandonment dataset. Secondly, we improved PSPNet by fusing statistical texture learning modules to learn multiple texture information on low-level feature maps, and combined high-level semantic features for cropland abandonment recognition. Experiments are performed on the HRAC dataset. Compared with other methods, the proposed model has the best performance on this dataset, both in terms of accuracy and visualization, proving that deep mining of low-level statistical texture features is beneficial for crop abandonment recognition.

Space Target Classification With Corrupted HRRP Sequences Based on Temporal–Spatial Feature Aggregation Network

High-resolution range profile (HRRP) sequences have great potential for space target classification because they can provide both scattering information and micromotion information. However, many factors cause an obtained HRRP sequence for a space target to be corrupted in real cases due to noise interference, limited radar resources, and the requirement of multitarget observations. Many space target classification methods cease to be effective when HRRP sequences are corrupted, so classifying space targets with corrupted HRRP sequences is still a challenging problem. To solve this problem, a novel space target classification method based on a temporal–spatial feature aggregation network (TSFA-Net) is proposed by using the corrupted HRRP sequences directly. First, a sequence-to-token module (S2T-module) is designed to extract low-level and fine-grained features from the raw inputs. Second, to effectively model the long-range dependencies among corrupted HRRP sequences and capture global representations without losing target local features, we propose a parallel and dual-branch block, i.e., a temporal–spatial feature aggregation block (TSFA-block), by combining a Transformer network and a convolutional neural network (CNN). Then, via progressively hierarchically stacking TSFA-blocks, a hierarchical temporal–spatial feature aggregation subnetwork (H-TSFA-subnetwork) is constructed to obtain the final temporal–spatial features. Finally, a token-to-label module (T2L-module) is adopted to obtain the classification results. Extensive experiments demonstrate that the proposed method achieves state-of-the-art classification accuracy for space target classification with HRRP sequences, especially under the conditions of a low signal-to-noise ratio and a high missing rate.

Experimental evaluation of unsupervised image retrieval application using hybrid feature extraction by integrating deep learning and handcrafted techniques

Abstract Content Based Image Retrieval is ever growing technology for many applications such as medical, remote sensing, social media search engines and surveillance monitoring, etc,. Representing image content with appropriate features is tedious task using traditional low level feature extraction methods. Deep learning models achieved high precision in classification and object detection algorithms by extracting automated high level feature extraction process. This paper proposed a hybrid feature extraction technique by combining the high level features and low level features to improve the robustness of the feature vector. The proposed model used pre-trained Googlenet model as feature extractor and combined with Gabor multiscale texture features. The final feature vector will be used for retrieving the relevant image data from the large scale image dataset. It has achieved the precision of 91 percent which shows better than state of art methods.

Novel Convolutions for Semantic Segmentation of Remote Sensing Images

The networks are required to be capable of learning low-level features well when applied to remote sensing image semantic segmentation task. To capture accurate and abundant low-level semantic information, the early feature extractor layer is crucial to the whole network, because all the subsequent features are inferred from that base. To address the low-level feature extraction issue and overcome the shortcomings of traditional convolution like too many parameters or limited receptive field, some novel convolution units have been proposed in the literature. In this paper, we propose two elaborately designed and portable yet effective convolution units, i.e., directional convolution (DC) and large field convolution (LFC), combined as the extractor of low-level semantic features. DC is designed to extract directional features from specific directions, and LFC can achieve a large receptive field with few parameters. Experiment results on two public datasets provide evidence that our convolution units can help deep learning networks improve performance stably and comprehensively compared to the baseline networks.

Early detection of liver disorders using hybrid soft computing techniques for optimal feature selection and classification

SummaryHuman liver disease is a sort of illness that starts in the ovaries and is particularly dangerous for women. As a consequence, aberrant cells develop that have the potential to spread to other parts of the body. Liver disease is a serious disorder that affects women's ovaries and is difficult to identify early on, which is why it is still one of the leading causes of mortality. The significance of unequivocal confirmation of intrinsic and typical components in establishing new frameworks to detect and eliminate danger is substantial. In this study, we present a hybrid soft computing approach for early diagnosis of liver problems (EDLD‐HS). For high and low level feature extraction models, we first employ an improved ant swarm optimization (IASO) approach. Then, for optimum feature selection, we develop a modified whale search optimization (MWSO) technique that combines features that may reflect both texture patterns and semantic backdrop scattered in the data. After that, we employed a hybrid swallow swarm intelligent‐deep neural network (HSSI‐DNN) classifier to determine the stage of liver illness. Finally, we used MATLAB R2014a to test our proposed EDLD‐HS method using well‐known benchmark datasets including BUPA, ILPD, and MPRLPD. The simulation findings are compared to existing state‐of‐the‐art methodologies in terms of accuracy, specificity, sensitivity, precision, recall, F1‐score, G‐mean, and area under curve (AUC). The detection accuracy of the proposed HSSI‐DNN classifier is 83.26% (BUPA), 84.26% (ILPD), and 91.23% (ILPD) (MPRLPD).

Temporal Convolutional Networks with RNN approach for chaotic time series prediction

The prediction of chaotic time series, which constitutes many systems in the field of science and engineering, has recently become the focus of attention of researchers. Chaotic time series prediction is making future predictions about these systems using previously observed data for a nonlinear chaotic system with a known initial condition. Chaotic time series prediction can be applied in many fields such as weather forecasting, finance and stock markets. Many disciplines work on solving time series prediction problem, ranging from forecasting weather events days in advance to traders predicting the future of stocks. In recent studies, it has been observed that hybrid deep neural network methods give better performance in solving time series prediction problems and have gained popularity in order to benefit from the advantages of more than one method in solving such problems. In this study, a hybrid deep neural network architecture is proposed for chaotic time series prediction. The used hybrid approach includes both temporal convolutional network to extract low level features from input and recurrent neural network layers such as long short-term memory and gated recurrent units to capture temporal information. Simulations were carried out on nine different chaotic time series dataset which are obtained from Lorenz, Rössler and a Lorenz-like chaotic equation sets, and twenty-one electrocardiogram (ECG) recordings of patients with arrhythmias. In the benchmark study, in which twelve different methods, including classical machine learning, deep neural network and hybrid models were used, the proposed model achieved the best prediction performance with an average root-mean-square error (RMSE) value of 0.0022 for chaotic dataset and 0.0082 for ECG arrhythmia dataset. Performance evaluation metrics show that the proposed hybrid architecture can compete with the models in state-of-the-art studies in chaotic time series prediction.

VGG-TSwinformer: Transformer-based deep learning model for early Alzheimer’s disease prediction

Background and objective: Mild cognitive impairment (MCI) is a transitional state between normal aging and Alzheimer’s disease (AD), and accurately predicting the progression trend of MCI is critical to the early prevention and treatment of AD. Brain structural magnetic resonance imaging (sMRI), as one of the most important biomarkers for the diagnosis of AD, has been applied in various deep learning models. However, due to the inherent disadvantage of deep learning in dealing with longitudinal medical image data, few applications of deep learning for longitudinal analysis of MCI, and the majority of existing deep learning algorithms for MCI progress prediction rely on the analysis of the sMRI images collected at a single time-point, ignoring the progressive nature of the disorder. Methods: In this work, we propose a VGG-TSwinformer model based on convolutional neural network (CNN) and Transformer for short-term longitudinal study of MCI. In this model, VGG-16 based CNN is used to extract low-level spatial features of longitudinal sMRI images and map these low-level features to high-level feature representations, sliding-window attention is used for fine-grained fusion of spatially adjacent feature representations, and gradually fuses distant spatial feature representations through the superposition of attention windows of different sizes, temporal attention is used to measure the evolution of this feature representations as a result of disease progression. Results: We validated our model on the ADNI dataset. For the classification task of sMCI vs pMCI, accuracy, sensitivity, specificity and AUC reached 77.2%, 79.97%, 71.59% and 0.8153 respectively. Compared with other cross-sectional studies also applied to sMRI, the proposed model achieved better results in terms of accuracy, sensitivity, and AUC. Conclusion: The proposed VGG-TSwinformer is a deep learning model for short-term longitudinal study of MCI, which can build brain atrophy progression model from longitudinal sMRI images, and improve diagnostic efficiency compared to algorithms using only cross-sectional sMRI images.

Cell Phenotype Classification Based on Joint of Texture Information and Multilayer Feature Extraction in DenseNet.

Cell phenotype classification is a critical task in many medical applications, such as protein localization, gene effect identification, and cancer diagnosis in some types. Fluorescence imaging is the most efficient tool to analyze the biological characteristics of cells. So cell phenotype classification in fluorescence microscopy images has received increased attention from scientists in the last decade. The visible structures of cells are usually different in terms of shape, texture, relationship between intensities, etc. In this scope, most of the presented approaches use one type or joint of low-level and high-level features. In this paper, a new approach is proposed based on a combination of low-level and high-level features. An improved version of local quinary patterns is used to extract low-level texture features. Also, an innovative multilayer deep feature extraction method is performed to extract high-level features from DenseNet. In this respect, an output feature map of dense blocks is entered in a separate way to pooling and flatten layers, and finally, feature vectors are concatenated. The performance of the proposed approach is evaluated on the benchmark dataset 2D-HeLa in terms of accuracy. Also, the proposed approach is compared with state-of-the-art methods in terms of classification accuracy. Comparison of results demonstrates higher performance of the proposed approach in comparison with some efficient methods.

An Improved Disc Segmentation Based on U-Net Architecture for Glaucoma Diagnosis

Various computer-aided diagnosis systems have been expanded and used for diagnosing glaucoma. Since the optic disc and optic cup are the main parameters for the early detection of glaucoma, this study proposes an accurate CAD system that firstly detects the optic disc and cup then classifies them into normal or abnormal. The U-Net architecture is employed. Despite its excellent segmentation performances, this model repeatedly extracts low-level features, which leads to redundant use of computational sources. To address these issues, a two-stage segmentation of the optic disc and cup was proposed. Firstly, a region of interest (ROI) is extracted from the fundus images by localizing and cutting the optic disc zone. Then, a U-Net model was built in order to obtain the refined segmentation. The public REFUGE dataset is adopted to validate proposed system. After a data augmentation step, an average accuracy of 0.97 and 0.96 for predicted OD cut off area and predicted original images respectively are obtained.

Anomaly detection in Internet of medical Things with Blockchain from the perspective of deep neural network

IoMT technology has many advantages in healthcare system, such as optimizing the medical service model, improving the efficiency of hospital operation and management, and improving the overall service level of the hospital. IoMT devices do not have a security authentication mechanism, and the trust between devices relies heavily on centralized third-party services. Blockchain can provide a secure interactive environment for the medical Internet of Things. However, security issues in the IoMT-Blockchain environment are also becoming increasingly prominent. Cyber-attacks targeting IoMT-Blockchain will not only compromise the security of IoT devices, but also seriously affect the security of the Internet. Therefore, how to detect abnormal traffic in the IoMT-Blockchain environment becomes particularly important. In this work, an abnormal traffic detection with deep neural network is designed for abnormal traffic detection in IoMT-Blockchain environment. First, this work proposes a feature extraction algorithm based on multi-model autoencoders. The algorithm processes the feature information in groups to reduce the complexity between traffic feature information. It builds a multi-model autoencoder to further extract fusion features between multi-model features. Second, to maximize use of traffic data information in detection network, this work proposes a multi-feature sequence anomaly detection algorithm. The algorithm extracts low-level fusion features and high-level temporal features in network traffic respectively, and applies the features to anomaly detection and classification tasks by means of residual learning.

Automatic Extraction of Corner Points from Aerial Images Using Point-Feature Operators and Hough Transform

Low-level feature extraction such as lines and points (i.e. corners), forms a fundamental step in digital photogrammetry and other fields. They supply the inputs for the photogrammetric orientation procedures; and they serve as an intermediate input for other processes such as object recognition. With the accumulation of knowledge, the research community is in a better position to develop new generations of smart algorithms and solutions that possess a new level of maturity and understanding for the underlying challenges of automation. To this end, this paper presents an innovative approach for corner point extraction that combines the outputs from classical point feature operators with Hough Transform to generate a better hypothesis for a corner point that can be used for applications in urban areas. In particular, extracted point features were used to guide line extraction in a local neighbourhood by Hough Transform. Then the corner points that will be obtained from lines intersection in this local neighbourhood will be compared with their nearby ones that were extracted by point feature operators. Based on passing a set of criteria, the intersection points from lines will replace the point feature as aset of potential corner points. Experimental findings show promising results of the proposed approach that raises the confidence level of the extracted corners and eliminating outliers.

TransCUNet: UNet cross fused transformer for medical image segmentation

Accurate segmentation of medical images is crucial for clinical diagnosis and evaluation. However, medical images have complex shapes, the structures of different objects are very different, and most medical datasets are small in scale, making it difficult to train effectively. These problems increase the difficulty of automatic segmentation. To further improve the segmentation performance of the model, we propose a multi-branch network model, called TransCUNet, for segmenting medical images of different modalities. The model contains three structures: cross residual fusion block (CRFB), pyramidal pooling module (PPM) and gated axial-attention, which achieve effective extraction of high-level and low-level features of images, while showing high robustness to different size segmentation objects and different scale datasets. In our experiments, we use four datasets to train, validate and test the models. The experimental results show that TransCUNet has better segmentation performance compared to the current mainstream segmentation methods, and the model has a smaller size and number of parameters, which has great potential for clinical applications.