Efficient Feature Extraction Research Articles

Drivable Road Region Segmentation in Real Time with High Precision using Deep Learning

Abstract: This paper presents a novel approach that addresses the challenging task of real-time drivable road region extraction in computer vision. Semantic segmentation, which involves accurately identifying and segmenting objects in real-time data, is a complex problem. However, deep learning has proven to be a powerful technique for achieving semantic segmentation by automatically identifying patterns without the need for explicit programming. To tackle this task, the paper proposes a fusion of the YOLO algorithm and UNET architecture, leveraging their respective strengths. The YOLO algorithm enables high-speed object detection, while the UNET architecture provides advantages in global location utilization, contextual understanding, and performance, even with limited training samples. Importantly, the proposed method is lightweight, making it suitable for deployment on embedded systems with limited computational power. To optimize memory usage and capture context at different scales, the system employs dilated convolutions for efficient feature extraction. The algorithm exhibits exceptional performance in accurately segmenting irregular objects and handles diverse input data types, including images and videos, in real-time. Overall, this paper contributes significantly to the advancement of computer vision technologies and offers a valuable solution for real-time drivable road region extraction. Its potential applications include addressing driving challenges and enhancing safety in autonomous vehicles and intelligent transportation systems.

Deep Autoencoder-Based Integrated Model for Anomaly Detection and Efficient Feature Extraction in IoT Networks

The intrusion detection system (IDS) is a promising technology for ensuring security against cyber-attacks in internet-of-things networks. In conventional IDS, anomaly detection and feature extraction are performed by two different models. In this paper, we propose a new integrated model based on deep autoencoder (AE) for anomaly detection and feature extraction. Firstly, AE is trained based on normal network traffic and used later to detect anomalies. Then, the trained AE model is employed again to extract useful low-dimensional features for anomalous data without the need for a feature extraction training stage, which is required by other methods such as principal components analysis (PCA) and linear discriminant analysis (LDA). After that, the extracted features are used by a machine learning (ML) or deep learning (DL) classifier to determine the type of attack (multi-classification). The performance of the proposed unified approach was evaluated on real IoT datasets called N-BaIoT and MQTTset, which contain normal and malicious network traffics. The proposed AE was compared with other popular anomaly detection techniques such as one-class support vector machine (OC-SVM) and isolation forest (iForest), in terms of performance metrics (accuracy, precision, recall, and F1-score), and execution time. AE was found to identify attacks better than OC-SVM and iForest with fast detection time. The proposed feature extraction method aims to reduce the computation complexity while maintaining the performance metrics of the multi-classifier models as much as possible compared to their counterparts. We tested the model with different ML/DL classifiers such as decision tree, random forest, deep neural network (DNN), conventional neural network (CNN), and hybrid CNN with long short-term memory (LSTM). The experiment results showed the capability of the proposed model to simultaneously detect anomalous events and reduce the dimensionality of the data.

HyFormer: Hybrid Grouping-Aggregation Transformer and Wide-Spanning CNN for Hyperspectral Image Super-Resolution

Hyperspectral image (HSI) super-resolution is a practical and challenging task as it requires the reconstruction of a large number of spectral bands. Achieving excellent reconstruction results can greatly benefit subsequent downstream tasks. The current mainstream hyperspectral super-resolution methods mainly utilize 3D convolutional neural networks (3D CNN) for design. However, the commonly used small kernel size in 3D CNN limits the model’s receptive field, preventing it from considering a wider range of contextual information. Though the receptive field could be expanded by enlarging the kernel size, it results in a dramatic increase in model parameters. Furthermore, the popular vision transformers designed for natural images are not suitable for processing HSI. This is because HSI exhibits sparsity in the spatial domain, which can lead to significant computational resource waste when using self-attention. In this paper, we design a hybrid architecture called HyFormer, which combines the strengths of CNN and transformer for hyperspectral super-resolution. The transformer branch enables intra-spectra interaction to capture fine-grained contextual details at each specific wavelength. Meanwhile, the CNN branch facilitates efficient inter-spectra feature extraction among different wavelengths while maintaining a large receptive field. Specifically, in the transformer branch, we propose a novel Grouping-Aggregation transformer (GAT), comprising grouping self-attention (GSA) and aggregation self-attention (ASA). The GSA is employed to extract diverse fine-grained features of targets, while the ASA facilitates interaction among heterogeneous textures allocated to different channels. In the CNN branch, we propose a Wide-Spanning Separable 3D Attention (WSSA) to enlarge the receptive field while keeping a low parameter number. Building upon WSSA, we construct a wide-spanning CNN module to efficiently extract inter-spectra features. Extensive experiments demonstrate the superior performance of our HyFormer.

Intelligent energy management strategy for hybrid electric vehicles using reinforcement learning

ABSTRACT Predicting the performance of a Hybrid EV (Electric Vehicle) is significant to alleviate the complications of insufficient fuel and other battery-related problems that affect the driving capacity. Many energy management techniques with ML (Machine Learning) and DL (Deep Learning) methodologies have been deployed in predicting the operation of vehicles based on energy sources and their features. Still, it is being noticed that prediction errors have increased along with the increase or decrease of input data. To solve the challenges addressed in existing studies, the proposed approach intends to use both ML and DL to predict faults from battery features that considerably manage the energy. It is accomplished through efficient feature extraction employed by Bi-LSTM (Bidirectional Long Short Term Memory), which extracts the best energy features required for predicting the defects. From the extracted input features collected hourly, daily, and monthly from the dataset, classification is done using the Modified XGBoost Algorithm with Empirical Loss function. Normal and Abnormal functioning of the Hybrid EV is classified effectively. Followed by that, the prediction of improper functioning is made with the assistance of the vehicle Id to maximise the vehicle’s performance through consideration of battery constraints. Experimentation is performed with MATLAB simulation, and the performance of the proposed system is evaluated using error metrics like RMSE (Root Mean Square Error), MAE (Mean Absolute Error), MSE (Mean Square Error), and accuracy. Efficiency is exhibited through the minimum loss obtained through the proposed work. From analysis of comparing LSTM method with proposed model, it is found that the LSTM method produced RMSE value of 1.055% which is 0.84% reduced in proposed model. On the other hand, the MAE of LSTM is 0.826% and the proposed model is decreased with 0.174%. Further, the MSE of LSTM is 1.113% and is 0.69% reduced in proposed model.

Spectral–Spatial Features Exploitation Using Lightweight HResNeXt Model for Hyperspectral Image Classification

Hyperspectral image classification is vital for various remote sensing applications; however, it remains challenging due to the complex and high-dimensional nature of hyperspectral data. This paper introduces a novel approach to address this challenge by leveraging spectral and spatial features through a lightweight HResNeXt model. The proposed model is designed to overcome the limitations of traditional methods by combining residual connections and cardinality to enable efficient and effective feature extraction from hyperspectral images, capturing both spectral and spatial information simultaneously. Furthermore, the paper includes an in-depth analysis of the learned spectral–spatial features, providing valuable insights into the discriminative power of the proposed approach. The extracted features exhibit strong discriminative capabilities, enabling accurate classification even in challenging scenarios with limited training samples and complex spectral variations. Extensive experimental evaluations are conducted on four benchmark hyperspectral data sets, the Pavia university (PU), Kennedy Space Center (KSC), Salinas scene (SA), and Indian Pines (IP). The performance of the proposed method is compared with the state-of-the-art methods. The quantitative and visual results demonstrate the proposed approach’s high classification accuracy, noise robustness, and computational efficiency superiority. The HResNeXt obtained an overall accuracy on PU, KSC, SA, and IP, 99.46%, 81.46%, 99.75%, and 98.64%, respectively. Notably, the lightweight HResNeXt model achieves competitive results while requiring fewer computational resources, making it well-suited for real-time applications.

A study on hand gesture recognition algorithm realized with the aid of efficient feature extraction method and convolution neural networks: design and its application to VR environment

A study on hand gesture recognition algorithm realized with the aid of efficient feature extraction method and convolution neural networks: design and its application to VR environment

POINT-WISE CLASSIFICATION OF HIGH-DENSITY UAV-LIDAR DATA USING GRADIENT BOOSTING MACHINES

Abstract. Point-wise classification of 3D point clouds is a challenging task in point cloud processing, whereas, in particular, its application to high-density point clouds needs special attention because a large number of point clouds affect computational efficiency negatively. Although deep learning based models have been gaining popularity in recent years and have reached state-of-the-art results in accuracy for point-wise classification, their requirements of the high number of training samples and computational resources make those models inefficient for high-density 3D point clouds. However, traditional machine learning classifiers require less training samples, so they are capable of reducing computational requirements, even considering the latest machine learning classifiers, particularly in ensemble learning of gradient boosting machines, the results can compete with deep learning models. In this study, we are studying the point-wise classification of high-density UAV LiDAR data and focusing on efficient feature extraction and a recent state-of-the-art gradient boosting machine learning classifier, LightGBM. Our proposed framework includes the following steps: at first, we are using point cloud sampling for creating sub-sampled point clouds, then we are calculating the features based on those scales implemented on GPU. Finally, we are using the LightGBM classifier for training and testing. For the evaluation of our framework, we used a publicly available benchmark dataset, Hessigheim 3D. According to the results, we achieved an overall accuracy of 87.59% and an average F1 score of 75.92%. Our framework has promising results and scores closer to deep learning models. However, more distinctive features are required to obtain more accurate results.

Soil classification using active contour model for efficient texture feature extraction

Soil classification using active contour model for efficient texture feature extraction

Development of an effective character segmentation and efficient feature extraction technique for malayalam character recognition from palm leaf manuscripts

Development of an effective character segmentation and efficient feature extraction technique for malayalam character recognition from palm leaf manuscripts

Hybrid CNN-Transformer model for medical image segmentation with pyramid convolution and multi-layer perceptron

Vision Transformer (ViT) has emerged as a potential alternative to convolutional neural networks for large datasets. However, applying ViT directly to medical image segmentation is challenging due to its lack of induction bias, which requires a large number of high-quality annotated medical images for effective model training. Recent studies have discovered that, in addition to the increased model capacity and generalization resulting from the lack of induction bias, the excellent performance of Transformer can also be attributed to its large receptive field. In this paper, we propose a U-shaped medical image segmentation model that combines large kernel convolutions with Transformers. Specifically, we construct a basic Transformer unit using pyramidal convolution modules with multi-scale kernels and multi-layer perceptron. In the pyramid convolution module, we employ grouped convolution to reduce parameter and computational complexity while utilizing multi-scale large kernel attention as a foundation for more efficient feature extraction. For different types of grouping, different sizes of convolutions are used to enhance the extraction of features with multiple receptive fields. To optimize the extracted features from the encoder, the U-shaped model integrates a variant of the pyramidal convolutional module into the skip connections. This variant utilizes multi-scale large kernel convolutional attention based on channel splitting. The incorporation of this variant enables efficient refinement of the feature representations within the skip connections. Through extensive comparisons on multi-modal medical image datasets, our model outperforms state-of-the-art methods across various evaluation metrics, with notable superiority observed on small-scale medical datasets. Our research findings suggest that the combination of large kernel convolutions and Transformer models introduces an advantageous inductive bias, resulting in enhanced performance specifically for small-scale medical image datasets. To facilitate accessibility, we have made our code openly accessible on our GitHub repository, which can be found at https://github.com/medical-images-process/CNN-Transformer.

Region-based Convolutional Neural Network Driven Alzheimer’s Severity Prediction

It's important to note that Alzheimer's disease can also affect individuals over the age of 60, and in fact, the risk of developing Alzheimer's increases with age. Additionally, while deep learning approaches have shown promising results in detecting Alzheimer's disease, they are not the only techniques available for diagnosis and treatment. That being said, using Region-based Convolutional Neural Network (RCNN) for efficient feature extraction and classification can be a valuable tool in detecting Alzheimer's disease. This new approach to identifying Alzheimer's disease could lead to a more accurate and personalized diagnosis. It can also help in early treatment and intervention. However, it's still important to continue developing new methods and techniques for this disorder. Considering this our work proposes an innovative Region-based Convolutional Neural Network Driven Alzheimer’s Severity Prediction approach in this paper. The exhaustive experimental result carried out, which proves the efficacy of our Alzheimer prediction system.

A practical wearable fall detection system based on tiny convolutional neural networks

Falls are a major public health problem in a rapidly aging society due to their high prevalence and severe consequences among the older population. Therefore, automatic fall detection is of primary importance to enable timely medical intervention for older fall victims. The existing fall detection methods suffer from either poor robustness or high computational cost, limiting their practical applications in wearable fall detection systems. In this study, a tiny convolutional neural network (TinyCNN) with two-stage efficient feature extraction was proposed and evaluated on two large-scale public fall datasets (KFall and SisFall) collected from wearable inertial sensors. Cross-validation results showed that TinyCNN achieved both mean sensitivities and specificities over 99 % on the KFall, and over 98 % on the SisFall. While benefiting from the effectiveness of TinyCNN, we also addressed its black-box effect using class activation map (CAM). Furthermore, by employing quantization techniques, TinyCNN achieved a low latency of 0.037 s on an ultra-low-power microcontroller (Arduino Nano BLE 33 Sense), enabling real-time performance without sacrificing the accuracy. Compared with existing state-of-the-art fall detection algorithms, the newly proposed TinyCNN yielded a more balanced performance in terms of accuracy and practicality. Inspired by these promising results, we further developed a practical wearable prototype system based on this embeddable TinyCNN and an accompanying mobile App for automatic fall detection and immediate alarm. This AI-enabled wearable system has great potential for real-life fall detection applications among older individuals.

Dataset classification: An efficient feature extraction approach for grammatical facial expression recognition

In this paper, an efficient features extraction using validated statistical approaches is proposed, along with a robust Grammatical Facial Expressions (GFEs) classifier in facial expression recognition systems. Accordingly, a new dataset was collected from 70 participants (33 males and 37 females) ranging in age from 18 to 46. The total number of video clips collected was 765.The features extracted in this study consist of 17 features associated with three categories of non-manual features: facial expression, head movement, and eye-gaze. Automatic recognition of nine classes of grammatical facial expressions in two languages (Arabic and Persian) is performed using a linear Support Vector Machine (SVM) classifier.The proposed system was also validated by testing it on the American Sign Language (ASL) dataset. In comparison to previous works on the ASL dataset, the results showed a higher accuracy rate of 95%.

A flexible deep learning framework for thermographic inspection of composites

Infrared thermography (IRT) is a promising inspection technique, showing good defect detectability in a range of materials. To advance the IRT inspection technique, automated defect detection and analysis are of high interest. This study proposes an object detection algorithm based on Faster R–CNN with attention-based feature fusion network and flexible late fusion strategy for efficient extraction of defect features from an IRT image sequence. In order to train the deep learning-based algorithm, a large, diverse and representative virtual thermographic dataset for composite materials was constructed by an in-house developed parameterized 3D finite element simulator. The virtually trained deep learning framework is tested on experimental IRT datasets which were obtained on composite coupons having a range of artificial defect types as well as on stiffened aerospace composite panels with real (production) defects. The obtained test results indicate the high performance, generalization and reliability of the proposed deep learning framework for automated thermographic inspection of composite parts.

BiTNet: A lightweight object detection network for real-time classroom behavior recognition with transformer and bi-directional pyramid network

In this paper, we propose BiTNet, a real-time object detection network, for enhancing teaching quality and providing feedback through real-time analysis of student behavior in classroom. BiTNet addresses challenges faced by current methods, such as occlusion and small objects in images. Specifically, it incorporates an Efficient Transformer Block (ETB) and Efficient Convolution Aggregation Block (ECAB) for extracting image features. ETB uses convolutional multi-head self-attention (CMHSA) to capture contextual information thus improves the capability to recoginze occluded objects. Its homogeneous multi-branch design reduces the computational costs. The single aggregation operation of ECAB aggregates the extracted features into a single feature map, improving the accuracy of recognition while reducing the computational cost. Then, the semantic and positional information extracted from the backbone are fused in Bi-directional Feature Pyramid Network (BiFPN). BiTNet also employs a smaller size detection head for recognition of small objects. Experimental results on classroom behavior dataset demonstrate that our proposed method has impressive accuracy (82.9%) and speed (256.7 FPS on GPU, 30.7 FPS on CPU, and 18.9 on camera). We further experiment with our method on Pascal VOC 2007 and VisDrone2021, validating BiTNet’s generalization and its capability to recognize small objects. In conclusion, BiTNet has a simple yet efficient network structure and feature extraction module, which reduce computational cost while maintaining recognition accuracy, making it a contender for real-time classroom behavior recognition tasks.

Flattening-Net: Deep Regular 2D Representation for 3D Point Cloud Analysis.

Point clouds are characterized by irregularity and unstructuredness, which pose challenges in efficient data exploitation and discriminative feature extraction. In this paper, we present an unsupervised deep neural architecture called Flattening-Net to represent irregular 3D point clouds of arbitrary geometry and topology as a completely regular 2D point geometry image (PGI) structure, in which coordinates of spatial points are captured in colors of image pixels. Intuitively, Flattening-Net implicitly approximates a locally smooth 3D-to-2D surface flattening process while effectively preserving neighborhood consistency. As a generic representation modality, PGI inherently encodes the intrinsic property of the underlying manifold structure and facilitates surface-style point feature aggregation. To demonstrate its potential, we construct a unified learning framework directly operating on PGIs to achieve diverse types of high-level and low-level downstream applications driven by specific task networks, including classification, segmentation, reconstruction, and upsampling. Extensive experiments demonstrate that our methods perform favorably against the current state-of-the-art competitors. The source code and data are publicly available at https://github.com/keeganhk/Flattening-Net.

Optimized deep belief network and unsupervised deep learning methods for disease prediction

Due to the vast amount of patient health data, automated healthcare systems still struggle to classify and diagnose various ailments. Learning redundant data also reduces categorization accuracy. A Deep Belief Network (DBN) has been used to precisely extract the most important aspects from clinical data by ignoring irrelevant/redundant features. Due of many learning variables, training is complicated. Similarly, the hybrid model has been employed by ensemble Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) to categorize diseases. But, its efficiency depends on the proper choice of kernels and hyper-parameters. Therefore, this paper develops an efficient feature extraction and classification model for healthcare systems. First, several medical data related to the patient’s health are collected. Then, an Optimized DBN (ODBN) model is presented for maximizing the accurateness of DBN by optimizing the learning variables depends on the Ant Lion Optimization (ALO) algorithm. With learning ODBN, the most relevant features are extracted with reduced computational complexity. After that, the CNN-LSTM with Unsupervised Fine-tuned Deep Self-Organizing Map (UFDSOM)-based classifier model is designed to categorize the extracted features into categories of illnesses. In this novel classifier, dropout normalization and parameter tuning processes are applied to avoid overfitting and optimize the hyper-parameters, which results in a less training period. In the end, studies utilizing publically accessible datasets show that the ODBN with CNN-LSTM-UFDSOM system outperforms classical models by 98.23%.

ECLPOD: An Extremely Compressed Lightweight Model for Pear Object Detection in Smart Agriculture

Accurate pear sorting plays a crucial role in ensuring the quality of pears and increasing the sales of them. In the domain of intelligent pear sorting, precise target detection of pears is imperative. However, practical implementation faces challenges in achieving adequate accuracy in pear target detection due to the limitations of computational resources in embedded devices and the occurrence of occlusion among pears. To solve this problem, we built an image acquisition system based on pear sorting equipment and created a pear dataset containing 34,598 pear images under laboratory conditions. The dataset was meticulously annotated using the LabelImg software, resulting in a total of 154,688 precise annotations for pears, pear stems, pear calyxes, and pear defects. Furthermore, we propose an Extremely Compressed Lightweight Model for Pear Object Detection (ECLPOD) based on YOLOv7’s pipeline to assist in the pear sorting task. Firstly, the Hierarchical Interactive Shrinking Network (HISNet) was proposed, which contributed to efficient feature extraction with a limited amount of computation and parameters. The Bulk Feature Pyramid (BFP) module was then proposed to enhance pear contour information extraction during feature fusion. Finally, the Accuracy Compensation Strategy (ACS) was proposed to improve the detection capability of the model, especially for identification of the calyces and stalks of pears. The experimental results indicate that the ECLPOD achieves 90.1% precision (P) and 85.52% mAP50 with only 0.58 million parameters and 1.3 GFLOPs of computation in the homemade pear dataset in this paper. Compared with YOLOv7, the number of parameters and the amount of computation for the ECLPOD are compressed to 1.5% and 1.3%, respectively. Compared with other mainstream methods, the ECLPOD achieves an optimal trade-off between accuracy and complexity. This suggests that the ECLPOD is superior to these existing approaches in the field of object detection for assisting pear sorting tasks with good potential for embedded device deployment.

Normative and mechanistic model of an adaptive circuit for efficient encoding and feature extraction

One major question in neuroscience is how to relate connectomes to neural activity, circuit function, and learning. We offer an answer in the peripheral olfactory circuit of the Drosophila larva, composed of olfactory receptor neurons (ORNs) connected through feedback loops with interconnected inhibitory local neurons (LNs). We combine structural and activity data and, using a holistic normative framework based on similarity-matching, we formulate biologically plausible mechanistic models of the circuit. In particular, we consider a linear circuit model, for which we derive an exact theoretical solution, and a nonnegative circuit model, which we examine through simulations. The latter largely predicts the ORN [Formula: see text] LN synaptic weights found in the connectome and demonstrates that they reflect correlations in ORN activity patterns. Furthermore, this model accounts for the relationship between ORN [Formula: see text] LN and LN-LN synaptic counts and the emergence of different LN types. Functionally, we propose that LNs encode soft cluster memberships of ORN activity, and partially whiten and normalize the stimulus representations in ORNs through inhibitory feedback. Such a synaptic organization could, in principle, autonomously arise through Hebbian plasticity and would allow the circuit to adapt to different environments in an unsupervised manner. We thus uncover a general and potent circuit motif that can learn and extract significant input features and render stimulus representations more efficient. Finally, our study provides a unified framework for relating structure, activity, function, and learning in neural circuits and supports the conjecture that similarity-matching shapes the transformation of neural representations.

Subject-Independent per Beat PPG to Single-Lead ECG Mapping

In this paper, a beat-based autoencoder is proposed for mapping photoplethysmography (PPG) to a single-lead electrocardiogram (single-lead ECG) signal. The main limiting factors represented in uncleaned data, subject dependency, and erroneous beat segmentation are regarded. The dataset is cleaned by a two-stage clustering approach. Rather than complete single–lead ECG signal reconstruction, a beat-based PPG-to-single-lead-ECG (PPG2ECG) conversion is introduced for providing a simple lightweight model that meets the computational capabilities of wearable devices. In addition, peak-to-peak segmentation is employed for alleviating errors in PPG onset detection. Furthermore, subject-dependent training is highlighted as a critical factor in training procedures because most existing work includes different beats/signals from the same subject’s record in both training and testing sets. So, we provide a completely subject-independent model where the testing subjects’ records are hidden in the training stage entirely, i.e., a subject record appears once either in the training or testing set, but testing beats/signals belong to records that never appear in the training set. The proposed deep learning model is designed for providing efficient feature extraction that attains high reconstruction quality over subject-independent scenarios. The achieved performance is about 0.92 for the correlation coefficient and 0.0086 for the mean square error for the dataset extracted/cleaned from the MIMIC II dataset.