Lightweight Neural Network Research Articles

Inferring Electrocardiography From Optical Sensing Using Lightweight Neural Network

Inferring Electrocardiography From Optical Sensing Using Lightweight Neural Network

LDMRes-Net: A Lightweight Neural Network for Efficient Medical Image Segmentation on IoT and Edge Devices.

In this study, we propose LDMRes-Net, a lightweight dual-multiscale residual block-based convolutional neural network tailored for medical image segmentation on IoT and edge platforms. Conventional U-Net-based models face challenges in meeting the speed and efficiency demands of real-time clinical applications, such as disease monitoring, radiation therapy, and image-guided surgery. In this study, we present the Lightweight Dual Multiscale Residual Block-based Convolutional Neural Network (LDMRes-Net), which is specifically designed to overcome these difficulties. LDMRes-Net overcomes these limitations with its remarkably low number of learnable parameters (0.072 M), making it highly suitable for resource-constrained devices. The model's key innovation lies in its dual multiscale residual block architecture, which enables the extraction of refined features on multiple scales, enhancing overall segmentation performance. To further optimize efficiency, the number of filters is carefully selected to prevent overlap, reduce training time, and improve computational efficiency. The study includes comprehensive evaluations, focusing on the segmentation of the retinal image of vessels and hard exudates crucial for the diagnosis and treatment of ophthalmology. The results demonstrate the robustness, generalizability, and high segmentation accuracy of LDMRes-Net, positioning it as an efficient tool for accurate and rapid medical image segmentation in diverse clinical applications, particularly on IoT and edge platforms. Such advances hold significant promise for improving healthcare outcomes and enabling real-time medical image analysis in resource-limited settings.

UAV-Taken Wind Turbine Image Dehazing With a Double-Patch Lightweight Neural Network

UAV-Taken Wind Turbine Image Dehazing With a Double-Patch Lightweight Neural Network

Monocular visual anti-collision method based on residual mixed attention for storage and retrieval machines

In the traditional manufacturing industry, the safe operation of storage and retrieval (S/R) machines is vital for efficient automated warehouse management. Recently, deep learning-based monocular vision approaches have emerged as a promising solution for collision avoidance in automated storage and retrieval systems (AS/RS). However, the scarcity of specific datasets and the need for high-performance, lightweight neural network models pose challenges in the AS/RS domain. To address these challenges, this paper introduces a novel workflow for implementing a monocular anti-collision system for S/R machines. A semi-supervised annotation method leveraging a pretrained clustering model is proposed to expedite image collection and classification, resulting in a tailored dataset. Furthermore, the NightMix technique is introduced, combining Grayscale Enhancement and Floating Grid Mix method to enhance dataset diversity and capacity. Notably, a residual attention awareness module is integrated into MobileNet V3 to enhance its feature processing capabilities. Finally, the proposed approaches are validated through extensive experiments involving various design choices. The experimental results demonstrate that our methods are not only computationally efficient but also highly competitive in terms of learning performance.

Efficient Mixed-Type Wafer Defect Pattern Recognition Based on Light-Weight Neural Network.

Wafer defect pattern recognition can help engineers improve the production process of semiconductor chips. In real industrial scenarios, the recognition of mixed-type wafer defects is difficult and the production scale of semiconductor wafers is large, which requires high accuracy and speed in wafer defect pattern recognition. This study proposes a light-weight neural network model to efficiently recognize mixed-type wafer defects. The proposed model is constructed via inverted residual convolution blocks with attention mechanisms and large kernel convolution downsampling layers. The inference speed of the inverted residual convolution block is fast, and the attention mechanism can enhance feature extraction capabilities. Large kernel convolutions help the network retain more important feature information during downsampling operations. The experimental results on the real Mixed-type WM38 dataset show that the proposed model achieves a recognition accuracy of 98.69% with only 1.01 M parameters. Compared with some popular high-performance models and light-weight models, our model has advantages in both recognition accuracy and inference speed. Finally, we deploy the model as a TensorRT engine, which significantly improves the inference speed of the model, enabling it to process more than 1300 wafer maps per second.

Efficient and Lightweight Neural Network for Hard Hat Detection

Electric power operation, as one of the key fields in the world, faces particularly prominent safety issues. Ensuring the safety of operators has become the most fundamental requirement in power operation. However, there are some safety hazards in power construction. These hazards are mainly due to weak safety awareness among staff and the failure to standardize the wearing of safety helmets. In order to effectively address this situation, technical means such as video surveillance technology and computer vision technology can be utilized to monitor whether staff are wearing helmets and provide timely feedback. Such measures will greatly enhance the safety level of power operation. This paper proposes an improved lightweight helmet detection algorithm named YOLO-M3C. The algorithm first replaces the YOLOv5s backbone network with MobileNetV3, successfully reducing the model size from 13.7 MB to 10.2 MB, thereby increasing the model’s detection speed from 42.0 frames per second to 55.6 frames per second. Then, the CA attention mechanism is introduced into the backbone network to enhance the feature extraction capability of the model. Finally, in order to further improve the detection recall rate and accuracy of the model, a knowledge distillation of the model was carried out. The experimental results show that, compared with the original YOLOv5s algorithm, the average accuracy of the improved YOLO-M3C algorithm is improved by 0.123, and the recall rate is the same. These results verify that the algorithm YOLO-M3C has excellent performance in target detection and recognition, which can improve accuracy and confidence, while reducing false detection and missing detection, and effectively meet the needs of helmet-wearing detection.

A Forest Fire Smoke Monitoring System Based on a Lightweight Neural Network for Edge Devices

Forest resources are one of the indispensable resources of the earth, which are the basis for the survival and development of human society. With the swift advancements in computer vision and artificial intelligence technology, the utilization of deep learning for smoke detection has achieved remarkable results. However, the existing deep learning models have poor performance in forest scenes and are difficult to deploy because of numerous parameters. Hence, we introduce an optimized forest fire smoke monitoring system for embedded edge devices based on a lightweight deep learning model. The model makes full use of the multi-scale variable attention mechanism of Transformer architecture to strengthen the ability of image feature extraction. Considering the needs of application scenarios, we propose an improved lightweight network model LCNet for feature extraction, which can reduce the parameters and enhance detecting ability. In order to improve running speed, a simple semi-supervised label knowledge distillation scheme is used to enhance the overall detection capability. Finally, we design and implement a forest fire smoke detection system on an embedded device, including the Jetson NX hardware platform, high-definition camera, and detection software system. The lightweight model is transplanted to the embedded edge device to achieve rapid forest fire smoke detection. Also, an asynchronous processing framework is designed to make the system highly available and robust. The improved model reduces three-fourths of the parameters and increases speed by 3.4 times with similar accuracy to the original model. This demonstrates that our system meets the precision demand and detects smoke in time.

UBO-EREX: Uncertainty Bayesian-Optimized Extreme Recurrent EXpansion for Degradation Assessment of Wind Turbine Bearings

Maintenance planning is crucial for efficient operation of wind turbines, particularly in harsh conditions where degradation of critical components, such as bearings, can lead to costly downtimes and safety threats. In this context, prognostics of degradation play a vital role, enabling timely interventions to prevent failures and optimize maintenance schedules. Learning systems-based vibration analysis of bearings stands out as one of the primary methods for assessing wind turbine health. However, data complexity and challenging conditions pose significant challenges to accurate degradation assessment. This paper proposes a novel approach, Uncertainty Bayesian-Optimized Extreme Recurrent EXpansion (UBO-EREX), which combines Extreme Learning Machines (ELM), a lightweight neural network, with Recurrent Expansion algorithms, a recently advanced representation learning technique. The UBO-EREX algorithm leverages Bayesian optimization to optimize its parameters, targeting uncertainty as an objective function to be minimized. We conducted a comprehensive study comparing UBO-EREX with basic ELM and a set of time-series adaptive deep learners, all optimized using Bayesian optimization with prediction errors as the main objective. Our results demonstrate the superior performance of UBO-EREX in terms of approximation and generalization. Specifically, UBO-EREX shows improvements of approximately 5.1460 ± 2.1338% in the coefficient of determination of generalization over deep learners and 5.7056% over ELM, respectively. Moreover, the objective search time is significantly reduced with UBO-EREX with 99.7884 ± 0.2404% over deep learners, highlighting its effectiveness in real-time degradation assessment of wind turbine bearings. Overall, our findings underscore the significance of incorporating uncertainty-aware UBO-EREX in predictive maintenance strategies for wind turbines, offering enhanced accuracy, efficiency, and robustness in degradation assessment.

MSSD-Net: Multi-Scale SAR Ship Detection Network

In recent years, the development of neural networks has significantly advanced their application in Synthetic Aperture Radar (SAR) ship target detection for maritime traffic control and ship management. However, traditional neural network architectures are often complex and resource intensive, making them unsuitable for deployment on artificial satellites. To address this issue, this paper proposes a lightweight neural network: the Multi-Scale SAR Ship Detection Network (MSSD-Net). Initially, the MobileOne network module is employed to construct the backbone network for feature extraction from SAR images. Subsequently, a Multi-Scale Coordinate Attention (MSCA) module is designed to enhance the network’s capability to process contextual information. This is followed by the integration of features across different scales using an FPN + PAN structure. Lastly, an Anchor-Free approach is utilized for the rapid detection of ship targets. To evaluate the performance of MSSD-Net, we conducted extensive experiments on the Synthetic Aperture Radar Ship Detection Dataset (SSDD) and SAR-Ship-Dataset. Our experimental results demonstrate that MSSD-Net achieves a mean average precision (mAP) of 98.02% on the SSDD while maintaining a compact model size of only 1.635 million parameters. This indicates that MSSD-Net effectively reduces model complexity without compromising its ability to achieve high accuracy in object detection tasks.

DSEUNet: A lightweight UNet for dynamic space grouping enhancement for skin lesion segmentation

A prevalent cancer, skin cancer, requires precise segmentation for effective diagnosis and treatment. Despite the availability of numerous effective, lightweight neural network models, there remains a significant gap in their application in resource-limited settings, particularly in mobile healthcare. This study aims to bridge this gap by developing a model combining high efficiency and robust performance in such environments. This study introduces the Dynamic Spatial Group Enhanced Network (DSEUNet), significantly reducing the computational loads and parameter quantities while ensuring competitive segmentation accuracy. Our essential contribution is developing a model that excels in segmentation accuracy and addresses computational and parameter efficiency challenges in mobile healthcare applications. In extensive trials on the ISIC2017 and ISIC2018 datasets, DSEUNet not only surpasses traditional models like UNet in accuracy but does so with far fewer parameters and reduced computational demand. This research contributes to developing more accessible and efficient tools for medical image analysis, particularly in resource-constrained environments. As far as we know, this is the first model with 23.306 KB parameters and 9.497 M FLOPs.

Intelligent image recognition using lightweight convolutional neural networks model in edge computing environment

In order to enhance the performance of intelligent image recognition, this study optimizes the image recognition model through lightweight convolutional neural networks (CNNs) and cloud computing technology. The study begins by introducing the relevant theories and models of edge computing (EC) and lightweight CNNs models. Next, this study focuses on optimizing traditional image recognition models. Finally, the effectiveness and reliability of the proposed model are experimentally validated. The experimental results indicate that, when recognizing 1000 images, the average recognition times per image on cloud servers and edge servers are 13.33 ms and 50.11 ms, respectively. Despite the faster speed of cloud servers, the performance of edge servers can be improved by stacking servers. When the number of edge servers reaches 4, their recognition speed surpasses that of the cloud server model. Additionally, comparing the latency and processing time between EC and cloud computing architectures, it is observed that, with an increase in the number of processed images, the average processing time per image in the EC architecture remains relatively stable and consistent. In contrast, the average processing time gradually increases in the cloud computing architecture. This indicates a significant impact of the number of images on the processing rate of the cloud computing architecture. Therefore, as the time gap in processing between cloud computing and EC increases, the advantages of the EC architecture become more apparent. This study’s significance lies in advancing the development of deep learning technology and providing possibilities for its widespread practical application. The contribution of this study lies in promoting the development of EC and lightweight neural network models, offering valuable references and guidance for practical applications in related fields.

SAL: Optimizing the Dataflow of Spin-based Architectures for Lightweight Neural Networks

SAL: Optimizing the Dataflow of Spin-based Architectures for Lightweight Neural Networks

Millimeter wave gesture recognition using multi-feature fusion models in complex scenes

As a form of body language, the gesture plays an important role in smart homes, game interactions, and sign language communication, etc. The gesture recognition methods have been carried out extensively. The existing methods have inherent limitations regarding user experience, visual environment, and recognition granularity. Millimeter wave radar provides an effective method for the problems lie ahead gesture recognition because of the advantage of considerable bandwidth and high precision perception. Interfering factors and the complexity of the model raise an enormous challenge to the practical application of gesture recognition methods as the millimeter wave radar is applied to complex scenes. Based on multi-feature fusion, a gesture recognition method for complex scenes is proposed in this work. We collected data in variety places to improve sample reliability, filtered clutters to improve the signal’s signal-to-noise ratio (SNR), and then obtained multi features involves range-time map (RTM), Doppler-time map (DTM) and angle-time map (ATM) and fused them to enhance the richness and expression ability of the features. A lightweight neural network model multi-CNN-LSTM is designed to gestures recognition. This model consists of three convolutional neural network (CNN) for three obtained features and one long short-term memory (LSTM) for temporal features. We analyzed the performance and complexity of the model and verified the effectiveness of feature extraction. Numerous experiments have shown that this method has generalization ability, adaptability, and high robustness in complex scenarios. The recognition accuracy of 14 experimental gestures reached 97.28%.

Classification of Driver Cognitive Load in Conditionally Automated Driving: Utilizing Electrocardiogram-Based Spectrogram with Lightweight Neural Network

With the development of conditionally automated driving, drivers will be allowed to perform non-driving-related tasks. Under such circumstances, continuous monitoring of driver cognitive load will play an increasingly important role in ensuring that drivers have sufficient mental resources to take over control of the vehicle should the driving automation fail. However, estimation of cognitive load is challenging because of the difficulties in identifying high-level feature representation and accounting for interindividual differences. Physiological measures are believed to be promising candidates for cognitive load estimation in partially automated vehicles. However, current estimation methods are mainly based on the manual feature extraction of time- or frequency-domain indicators from physiological signals, which may not adapt to dynamic driving conditions. With the development of deep learning, the neural network has shown good performance in automatically capturing high-level features from input data. Inspired by this, we adopted a novel approach to classify driver cognitive load based on electrocardiogram (ECG) spectrograms, in which the driver’s ECG signal was collected and transformed into a 2D spectrogram by a short-time Fourier transform. A squeeze-and-excitation network-based deep-learning framework that can capture high-level features and pays more attention to the cognition-related features of the spectrogram was proposed for classification. Experiments on a publicly available dataset demonstrated that our model achieved an accuracy of 96.76% in differentiating two levels of cognitive load for a within-subject evaluation and 71.50% accuracy with an across-subjects evaluation. The results demonstrated the feasibility of detecting drivers’ cognitive load through deep learning using ECG spectrogram alone.

Methodology for Severe Convective Cloud Identification Using Lightweight Neural Network Model Ensembling

This study introduces an advanced ensemble methodology employing lightweight neural network models for identifying severe convective clouds from FY-4B geostationary meteorological satellite imagery. We have constructed a FY-4B based severe convective cloud dataset by a combination of algorithms and expert judgment. Through the ablation study of a model ensembling combination of multiple specialized lightweight architectures—ENet, ESPNet, Fast-SCNN, ICNet, and MobileNetV2—the optimal EFNet (ENet- and Fast-SCNN-based network) not only achieves real-time processing capabilities but also ensures high accuracy in severe weather detection. EFNet consistently outperformed traditional, heavier models across several key performance indicators: achieving an accuracy of 0.9941, precision of 0.9391, recall of 0.9201, F1 score of 0.9295, and computing time of 18.65 s over the test dataset of 300 images (~0.06 s per 512 × 512 pic). ENet shows high precision but misses subtle clouds, while Fast-SCNN has high sensitivity but lower precision, leading to misclassifications. EFNet’s ensemble approach balances these traits, enhancing overall predictive accuracy. The ensemble method of lightweight models effectively aggregates the diverse strengths of the individual models, optimizing both speed and predictive performance.

PHNNs: Lightweight Neural Networks via Parameterized Hypercomplex Convolutions.

Hypercomplex neural networks have proven to reduce the overall number of parameters while ensuring valuable performance by leveraging the properties of Clifford algebras. Recently, hypercomplex linear layers have been further improved by involving efficient parameterized Kronecker products. In this article, we define the parameterization of hypercomplex convolutional layers and introduce the family of parameterized hypercomplex neural networks (PHNNs) that are lightweight and efficient large-scale models. Our method grasps the convolution rules and the filter organization directly from data without requiring a rigidly predefined domain structure to follow. PHNNs are flexible to operate in any user-defined or tuned domain, from 1-D to [Formula: see text] regardless of whether the algebra rules are preset. Such a malleability allows processing multidimensional inputs in their natural domain without annexing further dimensions, as done, instead, in quaternion neural networks (QNNs) for 3-D inputs like color images. As a result, the proposed family of PHNNs operates with 1/n free parameters as regards its analog in the real domain. We demonstrate the versatility of this approach to multiple domains of application by performing experiments on various image datasets and audio datasets in which our method outperforms real and quaternion-valued counterparts. Full code is available at: https://github.com/eleGAN23/HyperNets.

ConvEQ: Convolutional neural network for earthquake phase classification using short time frequency transform

We present ConvEQ as a tool for discriminating seismic phases, leveraging artificial intelligence technique (Convolutional Neural Network) for short-time Frequency Transform of the seismic signal. Timely detection of the vertical (P) wave from an earthquake can generate a warning several tens of precious seconds before the more destructive waves strike. We propose a train-for-each-station approach for an Internet-of-Things-based Smart Earthquake Early Warning System, where lightweight neural networks trained for the seismic data belonging to each station are implemented on edge devices directly interfaced with seismometers. The approach has the potential to get the most from the sparse seismic network for Pakistan and other third-world countries. We train networks for multi-station and single-station data and achieve 96% and 99% accuracy, respectively, proving that train-for-each-station maximizes accuracy. The total processing time (including preprocessing and inference) is about 30ms for each event, thus suitable for real-time deployment. We further compare the performance of ConvEQ on simulated real-time data with several state-of-the-art contemporary algorithms. Our proposed approach demonstrates a robust response on diverse metrics. The ConvEQZ classifies the vertical seismic signal component with high accuracy and the ConvEQX can classify any seismic data component, inculcating robustness against connectivity issues.

Hot rolled steel surface defect detection and classification using an automatic ensemble approach

This study introduces an ensemble-based Deep Neural Network (DNN) model for detecting defects on steel surfaces. The method suggested in this study classifies steel surface conditions into six possible fault categories, namely, crazing, inclusion, rolled in, pitted surface, scratches, and patches. The images undergo preprocessing and extraction of features in spatial and frequency domains using image segmentation techniques such as grey level difference method (GLDM), fast Fourier Transform (FFT), grey level co-occurrence matrix (GLCM), texture analysis and discrete wavelet transform (DWT). The ensembling of image features into a fused feature pool is carried out after the preprocessing of input images that are provided as input to a light-weight neural network model for training and testing. The performance of the model is comprehensively evaluated via an ablation study both before and after ensembling. In addition, the model capability is effectively analyzed using receiver operating characteristics (ROC) curve, confusion matrix from which classification accuracy of the model could be obtained and other parameters including precision and f1-score. It was observed that the proposed deep learning network presents phenomenally high accuracy of 99.72% for detection and classification of steel surface faults. This result was found to be superior when compared with the performance of the same neural network over each feature type individually. This study also compares the classification results of the model built based on the ensembled feature set with the results of various other classification approaches available in literature. The ensemble-based model could potentially be integrated into existing inspection systems for real-time, efficient and robust condition monitoring of steel surfaces.

Multicore embedded sensing system based on lightweight neural network

ABSTRACT To the complexity of networks and the diversity of circuits, multicore embedded sensing systems suffer from low accuracy and efficiency in measuring temperature. To improve the measurement accuracy and efficiency of multicore embedded sensing systems, this paper utilised knowledge distillation, model pruning and parameter quantisation to lightweight neural networks. Meanwhile, the lightweight neural network was applied to multicore embedded sensing systems and the layout of multicore embedded sensing systems based on it was analysed from the perspectives of processor layout, storage design and link network, providing a reference and theoretical basis for further application of multicore embedded sensing systems.

Building Extraction from LiDAR Point Clouds Based on Revised RandLA-Net

Abstract. 3D building models is crucial for applications in smart cities. Automatic reconstruction of 3D buildings has been investigated based on various data sources. Point clouds from airborne LiDAR scanners can be used to extract buildings data due to its high accuracy and point density. In this paper, we present a methodology to segment buildings and corresponding rooftop structure from point clouds. First, RandLA-Net, which is an efficient and lightweight neural network for semantic segmentation of large-scale point clouds, is revised and adopted for building segmentation. By implementing local feature aggregation of each point, RandLA-Net can effectively preserve geometric details in point clouds. Besides 3D coordinates of point clouds, we incorporated point attributes including pulse intensity and return numbers into the network as additional features. Feature normalizations are applied to the input features. To achieve a better result of the local feature aggregation, hyperparameters of the network are fine-tuned according to the density of points and building size. Based on the classified building point clouds, DBSCAN clustering algorithm is implemented for segmenting individual buildings. Elevation histogram analysis is conducted to determine optimal threshold values for delineating candidate rooftop point clouds of individual buildings. For the buildings with multiple rooftops, multiple elevation threshold values are necessary to extract corresponding rooftops or walls. Then DBSCAN is employed again for segmentation of individual rooftops and denoising of point clouds of each building. Finally, Alpha-shape analysis is applied based on adaptive threshold values to build the envelope of each rooftop. Experiments show that our implementation of building segmentation using RandLA-net achieves higher mean IoU (Intersection over Union) and better classification performance in building segmentation. ISPRS benchmark data was used in our experiment and our methodology produce results with accuracy of 90.79%.