Light CNN Research Articles

LCADNet: a novel light CNN architecture for EEG-based Alzheimer disease detection.

Alzheimer's disease (AD) is a progressive and incurable neurologi-cal disorder with a rising mortality rate, worsened by error-prone, time-intensive, and expensive clinical diagnosis methods. Automatic AD detection methods using hand-crafted Electroencephalogram (EEG) signal features lack accuracy and reliability. A lightweight convolution neural network for AD detection (LCADNet) is investigated to extract disease-specific features while reducing the detection time. The LCADNet uses two convolutional layers for extracting complex EEG features, two fully connected layers for selecting disease-specific features, and a softmax layer for predicting AD detection probability. A max-pooling layer interlaced between convolutional layers decreases the time-domain redundancy in the EEG signal. The efficiency of the LCADNet and four pre-trained models using transfer learning is compared using a publicly available AD detection dataset. The LCADNet shows the lowest computation complexity in terms of both the number of floating point operations and inference time and the highest classification performance across six measures. The generalization of the LCADNet is assessed by cross-testing it with two other publicly available AD detection datasets. It outperforms existing EEG-based AD detection methods with an accuracy of 98.50%. The LCADNet may be a valuable aid for neurologists and its Python implemen- tation can be found at github.com/SandeepSangle12/LCADNet.git.

Novel Light Convolutional Neural Network for COVID Detection with Watershed Based Region Growing Segmentation.

A rapidly spreading epidemic, COVID-19 had a serious effect on millions and took many lives. Therefore, for individuals with COVID-19, early discovery is essential for halting the infection's progress. To quickly and accurately diagnose COVID-19, imaging modalities, including computed tomography (CT) scans and chest X-ray radiographs, are frequently employed. The potential of artificial intelligence (AI) approaches further explored the creation of automated and precise COVID-19 detection systems. Scientists widely use deep learning techniques to identify coronavirus infection in lung imaging. In our paper, we developed a novel light CNN model architecture with watershed-based region-growing segmentation on Chest X-rays. Both CT scans and X-ray radiographs were employed along with 5-fold cross-validation. Compared to earlier state-of-the-art models, our model is lighter and outperformed the previous methods by achieving a mean accuracy of 98.8% on X-ray images and 98.6% on CT scans, predicting the rate of 0.99% and 0.97% for PPV (Positive predicted Value) and NPV (Negative predicted Value) rate of 0.98% and 0.99%, respectively.

Type 2 Diabetes Detection With Light CNN From Single Raw PPG Wave

Type 2 Diabetes Detection With Light CNN From Single Raw PPG Wave

Testing automation adoption influencers in construction using light deep learning

Technology adoption is pivotal for the productivity growth in construction industry. This research paper attempts to fill this gap by addressing the following research objectives. First, the predictor factors stimulating project managers' adoption of construction automation innovations are rigorously analyzed using a mixed approach combining a systematic literature review, a knowledgeable panel, and a survey questionnaire. Secondly, the study implements a light deep learning model to track the progress of reinforcing bar placements by verifying completed rebar ties. By linking the progress of the bar to a single binary condition, the number of classes needed to train the neural network drops to only two resulting in a light CNN with a recall rate of 89.2% and precision rate of 95.7%. This model can be implemented on a low power GPU, making it more cost efficient and simpler to adopt on site. A similar approach can be used on other critical activities in construction. This approach can aid inspections, quality control, in combination with drones or robotic systems. The proposed system integrates the most important factors of a successful adoption by providing a proof of concept with potential use cases in construction sites.

Eye-referenced dynamic bounding box for face recognition using light convolutional neural network

Face recognition systems have imprinted its presence in many applications from offices to security to daily use as in personal digital gadgets. With many Face recognition systems in use, still there is scope for its performance improvement. The performance of such systems suffers due to presence of covariates like non-uniform illumination, pose deviations, occlusions, low resolution etc. Extracting the region of interest from input face images is very crucial steps in face recognition system. We proposed methodology to create a bounding box for every sample face image. This bounding box dynamically tries to fit the face image to cover the relevant face features useful for recognition purpose. We used a light CNN structure to perform the experiment on MUCT face dataset using three different methodologies for extracting region of interests from sample face images using bounding box. These are 1) min-max based ROI 2) OpenCV face alignment and 3) eye-reference based ROI. It is observed that the proposed eye-reference based face alignment method works better than conventional methods of min-max based ROI and OpenCV face alignment with considerable amount of improvement in the recognition accuracy. Our future work includes use of other structured datasets with various covariates for face recognition.

On significance of constant-Q transform for pop noise detection

Liveness detection has emerged as an important research issue for many biometrics, such as face, iris, hand geometry, etc. and significant research efforts are reported in the literature. However, less emphasis is given to liveness detection for voice biometrics or Automatic Speaker Verification (ASV). Voice Liveness Detection (VLD) can be a potential technique to detect spoofing attacks in ASV system. Presence of pop noise in the speech signal of live speaker provides the discriminative acoustic cue to distinguish between genuine vs. spoofed speech in the framework of VLD. Pop noise comes out as a burst at the lips, which is captured by the ASV system (since the speaker and microphone are close enough), indicating the liveness of the speaker and provides the basis of VLD. In this paper, we present the Constant-Q Transform (CQT) -based approach over the traditional Short-Time Fourier Transform (STFT) -based algorithm (baseline). With respect to Heisenberg’s uncertainty principle in signal processing framework, the CQT has variable spectro-temporal resolution, in particular, better frequency resolution for low frequency region and better temporal resolution for high frequency region, which can be effectively utilized to identify the low frequency characteristics of pop noise. We have also compared proposed algorithm with cepstral features, namely, Linear Frequency Cepstral Coefficients (LFCC) and Constant-Q Cepstral Coefficients. The experiments are performed on recently released POp noise COrpus (POCO) dataset with various statistical, discriminative, and deep learning-based classifiers, namely, Gaussian Mixture Model (GMM), Support Vector Machine (SVM), Convolutional Neural Networks (CNN), Light-CNN (LCNN), and Residual Network (ResNet), respectively. The significant improvement in performance, in particular, an absolute improvement of 14.23% and 10.95% in terms of percentage classification accuracy on development and evaluation set, respectively, is obtained for the proposed CQT-based algorithm along with SVM classifier, over the STFT-SVM (baseline) system. Similar trend of the performance improvement is observed for the GMM, CNN, LCNN, and ResNet classifiers for the proposed CQT-based algorithm vs. traditional STFT-based algorithm. The analysis is further extended by simulating the replay mechanism (in the standard framework of ASVSpoof-2019 PA challenge dataset) on the subset of POCO dataset in order to observe the effect of room acoustics onto the performance of the VLD system. By embedding the moderate simulated replay mechanism in POCO dataset, we obtained the percentage classification accuracy of 97.82% on evaluation set.

DeepFake Video Detection through Facial Sparse Optical Flow based Light CNN

DeepFake detection has become an attractive research topic with tremendous growth of interests recently. However, existing DeepFake detection studies spare no effort to improve accuracy or Area Under Curve metric, regardless of computing costs. In this work, the tradeoff between result accuracy and computing resources is taken into consideration. A facial sparse optical flow method is proposed to extract spatio-temporal features representing facial expression incoherence, which helps to distinguish fake videos and real videos. The features fed into a light CNN model are highly compact and low-dimensional. The proposed method has an amazing small amount of parameters with high training speed and low usage of GPU memory. The low resource requirement makes it possible to port to embedded development platform.

Type 2 Diabetes Detection with Light CNN from Single Raw PPG Wave

Type 2 Diabetes Detection with Light CNN from Single Raw PPG Wave

심장 소리 분류를 위한 Inverted Residuals 기반 경량화 모델

For the treatment and prevention of heart diseases, which is the leading cause of high mortality rate globally, the need for healthcare devices equipped with artificial intelligence(AI) model that can monitor in real-time and analyze heart conditions is increasing. Therefore, in this study, we propose a light CNN that can be applied to healthcare devices, using the PASCAL data. The proposed model used MFCC feature extraction method suitable for heart sound range, The light CNN was designed with the inverted residuals used in MobileNetV2. The experiments showed that the proposed model with fewer 82.5% of the learnable parameters, achieved similar performance in accuracy within the range of 1 to 2% compared to the previous studies. It was confirmed that the proposed light CNN can be feasibly incorporated on mobile devices by means of comparative experiments in a reasonable amount of computation.

Facial Recognition Using a Lightweight Deep Neural Networks

Current facial recognition systems are still far away from the capability of the human’s face perception. Facial recognition systems can continue to be improved as technology evolves. The task of face recognition has been heavily explored in recent years. In this research, we provide our initial idea in developing Lightweight Deep Neural Networks for facial recognition. Although our goal was to create an optimal model that would exceed current facial recognition model performance, we could experiment and discover alternative approaches to multi-class facial recognition/classification. We tested with a dataset of 2800 images of men and women with specified image sizes. We created three CNN with various architectures, which we used to train with the chosen dataset for 20, 50, 100, and 200 classes per model. The experimental results exhibit the challenges of increasing the complexity of neural networks. From these results, we concluded that a Light CNN Model with a small number of layers had an average test accuracy of 94.19%, which was the best classification performance on unseen data.

Scale fusion light CNN for hyperspectral face recognition with knowledge distillation and attention mechanism

Scale fusion light CNN for hyperspectral face recognition with knowledge distillation and attention mechanism

Modified Magnitude-Phase Spectrum Information for Spoofing Detection

Most of the existing feature representations for spoofing countermeasures consider information either from the magnitude or phase spectrum. We hypothesize that both magnitude and phase spectra can be beneficial for spoofing detection (SD) when collectively used to capture the signal artifacts. In this work, we propose a novel feature referred to as modified magnitude-phase spectrum (MMPS) to capture both magnitude and phase information from the speech signal. The constant-Q transform is used to obtain the magnitude and phase information in terms of MMPS, which can be denoted as CQT-MMPS. We then use this information for the proposal of a handcrafted feature, namely, constant-Q modified octave coefficients (CQMOC). To evaluate the proposed CQT-MMPS and CQMOC features, three classic anti-spoofing models are adopted, including the Gaussian mixture model (GMM), the light CNN (LCNN) and the ResNet. Additionally, since there is usually no prior knowledge about the spoofing kind in real-world applications, two novel methods referred to as three-class classifiers with maximum spoofing-score (TCMS) and multi-task learning (MTL) are designed for unknown-kind SD (UKSD). The experimental results on ASVspoof 2019 corpus show that CQMOC outperforms most of the commonly-used handcrafted features, and the CQT-based MMPS performs better than the magnitude-phase spectrum and the commonly-used log power spectrum. Further, the MMPS-based systems can achieve comparable or even better performance when compared with the state-of-the-art systems. We find that the newly-designed TCMS and MTL methods outperform the combination-based method for UKSD and meanwhile, generalize much better than the respective-kind-based methods in cross-spoofing-kind evaluation scenarios.

Multilevel fusing paired visible light and near-infrared spectral images for face anti-spoofing

Face anti-spoofing has become a vital element for guaranteeing the security of face recognition systems. Previous face anti-spoofing approaches generally exploit cues in visible light images or near-infrared images individually. Few studies pay attention to fusing visible light and near-infrared images for face anti-spoofing. However, the strengths and weaknesses of visible light images and near-infrared images for face anti-spoofing can be complementary. In this study, we introduce a new face anti-spoofing dataset named as CIGIT-PPM, which includes paired visible light and near-infrared images with spoofing medium, distance, pose, expression and session variations for both print and 3D mask attacks. Further, we propose a novel fusing paired visible light and near-infrared spectral images CNN based approach for face anti-spoofing, combining the discriminating ability of both visible light and near-infrared spectral images. Specifically, an end-to-end CNN is employed to learn fusing representation from paired images and do classification at multilevel, and then weight averaged strategy is utilized to integrate the classification probability of different levels, which gives the final result that the input paired images are captured from a live face or a spoof face. Extensive experiments show that our method achieves state-of-the-art results on self-collected dataset CIGIT-PPM and public dataset msspoof.

An embedded implementation of CNN-based hand detection and orientation estimation algorithm

Hand detection is an essential step to support many tasks including HCI applications. However, detecting various hands robustly under conditions of cluttered backgrounds, motion blur or changing light is still a challenging problem. Recently, object detection methods using CNN models have significantly improved the accuracy of hand detection yet at a high computational expense. In this paper, we propose a light CNN network, which uses a modified MobileNet as the feature extractor in company with the SSD framework to achieve robust and fast detection of hand location and orientation. The network generates a set of feature maps of various resolutions to detect hands of different sizes. In order to improve the robustness, we also employ a top-down feature fusion architecture that integrates context information across levels of features. For an accurate estimation of hand orientation by CNN, we manage to estimate two orthogonal vectors’ projections along the horizontal and vertical axes and then recover the size and orientation of a bounding box exactly enclosing the hand. In order to deploy the detection algorithm on embedded platform Jetson TK1, we optimize the implementations of the building modules in the CNN network. Evaluated on the challenging Oxford hand dataset, our method (the code is available at https://github.com/yangli18/hand_detection ) reaches 83.2% average precision at 139 FPS on a NVIDIA Titan X, outperforming the previous methods both in accuracy and efficiency. The embedded implementation of our algorithm has reached the processing speed of 16 FPS, which basically meets the requirement of real-time processing.

A Light CNN for End-to-End Car License Plates Detection and Recognition

License Plate Recognition (LPR) is of great significance due to its wide range of applications in the Intelligent Transportation System (ITS). It is an important and challenging research topic in image recognition fields. However, many of the current methods are still not robust in real-world complex scenario. The main contribution of this paper is to propose a multi-task convolutional neural network for license plate detection and recognition (MTLPR) with better accuracy and lower computational cost, and introduce a comprehensive data set of Chinese license plate. First, we train a Multi-task Convolutional Neural Networks (MTCNN) to detect license plate. Then we introduce an end-to-end method to recognize license plate information, which further improves the recognition precision. Last, We compare the experimental result with other state-of-the-art methods. The experimental result shows that our method achieves up to 98% recognition precision and is superior to other methods in the precision and speed of detection and recognition.

A Normalized Light CNN for Face Recognition

In recent years, achievements of deep learning in face recognition have exceeded those of traditional machine learning methods do. To achieve better recognition performance, the architectures of the neural network tend to be deeper and more complex, which wastes the time and space. Aiming at this issue, a series of light convolutional neural network (Light CNN) has been proposed, which consume less hardware resources while has excellent performance. In this paper, we propose a Normalized Light CNN model modified from a Light CNN model. We introduced a normalized layer to this model in both training and test phase. The normalized layer normalizes the output features, making it represent images better. We evaluate our model on LFW dataset. The accuracy of face verification reaches 98.46%, which is better than the original model.

Unconstrained face identification using maximum likelihood of distances between deep off-the-shelf features

The paper deals with unconstrained face recognition task for the small sample size problem based on computation of distances between high-dimensional off-the-shelf features extracted by deep convolution neural network. We present the novel statistical recognition method, which maximizes the likelihood (joint probabilistic density) of the distances to all reference images from the gallery set. This likelihood is estimated with the known asymptotically normal distribution of the Kullback–Leibler discrimination between nonnegative features. Our approach penalizes the individuals if their feature vectors do not behave like the features of observed image in the space of dissimilarities of the gallery images. We provide the experimental study with the LFW (Labeled Faces in the Wild), YTF (YouTube Faces) and IJB-A (IARPA Janus Benchmark A) datasets and the state-of-the-art deep learning-based feature extractors (VGG-Face, VGGFace2, ResFace-101, CenterFace and Light CNN). It is demonstrated, that the proposed approach can be applied with traditional distances in order to increase accuracy in 0.3–5.5% when compared to known methods, especially if the training and testing images are significantly different.

Face biometric quality assessment via light CNN

In this paper, we proposed a novel biometric quality assessment (BQA) method for face images and explored its applications in face recognition. Here, we considered five categories of common homogeneous distortion in video suvillance applications, i.e. low-resolution, blurring, additive Gaussian white noise, salt and pepper noise, and Poisson noise. In the BQA model, we first learnt a classifier to simultaneously predict the categories and degrees of the degradation in a face image. Because the quality labels are often ambiguous and inaccurate, we used a light convolutional neural network with the Max-Feature-Map units to make the BQA model robust to noisy labels. Afterwards, we calculated the biometric quality score by pooling such predictions based on the recognition confidence of each degradation class. Finally, we proposed one promising strategy for developing reliable face recognition systems based on this BQA method. Thorough experiments have been conducted on the CASIA, FLW, and YouTube databases. The results demonstrate the effectiveness of the proposed BQA method.