Convolutional Neural Network Framework Research Articles

A Novel Light Weight CNN Framework Integrated with Marine Predator Optimization for the Assessment of Tear Film-Lipid Layer Patterns

Tear film, the outermost layer of the eye, is a complex and dynamic structure responsible for tear production. The tear film lipid layer is a vital component of the tear film that provides a smooth optical surface for the cornea and wetting the ocular surface. Dry eye syndrome (DES) is a symptomatic disease caused by reduced tear production, poor tear quality, or excessive evaporation. Its diagnosis is a difficult task due to its multifactorial etiology. Out of several clinical tests available, the evaluation of the interference patterns of the tear film lipid layer forms a potential tool for DES diagnosis. An instrument known as Tearscope Plus allows the rapid assessment of the lipid layer. A grading scale composed of five categories is used to classify lipid layer patterns. The reported work proposes the design of an automatic system employing light weight convolutional neural networks (CNN) and nature inspired optimization techniques to assess the tear film lipid layer patterns by interpreting the images acquired with the Tearscope Plus. The designed framework achieves promising results compared with the existing state-of-the-art techniques.

Ascertaining Speech Emotion using Attention-based Convolutional Neural Network Framework

Ascertaining Speech Emotion using Attention-based Convolutional Neural Network Framework

Cascade Windows-Based Multi-Stream Convolutional Neural Networks Framework for Early Detecting In-Sleep Stroke Using Wristbands

Cascade Windows-Based Multi-Stream Convolutional Neural Networks Framework for Early Detecting In-Sleep Stroke Using Wristbands

A Multiscale Graph Convolutional Neural Network Framework for Fault Diagnosis of Rolling Bearing

A Multiscale Graph Convolutional Neural Network Framework for Fault Diagnosis of Rolling Bearing

Healthcare Monitoring Using Ensemble Classifiers in Fog Computing Framework

Nowadays, the cloud environment faces numerous issues like synchronizing information before the switch over the data migration. The requirement for a centralized internet of things (IoT)-based system has been restricted to some extent. Due to low scalability on security considerations, the cloud seems uninteresting. Since healthcare networks demand computer operations on large amounts of data, the sensitivity of device latency evolved among health networks is a challenging issue. In comparison to cloud domains, the new paradigms of fog computing give fresh alternatives by bringing resources closer to users by providing low latency and energy-efficient data processing solutions. Previous fog computing frameworks have various flaws, such as overvaluing response time or ignoring the accuracy of the result yet handling both at the same time compromises the network community. In this proposed work, Health Fog is integrated with the Optimized Cascaded Convolution Neural Network framework for diagnosing heart disease. Initially, the data is collected, and then pre-processing is done by Linear Discriminant Analysis. Then the features are extracted and optimized using Galactic Swarm Optimization. The optimized features are given into the Health Fog framework for diagnosing heart disease patients. It uses ensemble-based deep learning in edge computing devices, which automatically monitors real-life health networks such as heart disease analysis. Finally, the classifiers such as bagging, boosting, XGBoost, Multi-Layer Perceptron (MLP), and Partitions (PART) are used for classifying the data. Then the majority voting classifier predicts the result. This work uses FogBus architecture and evaluates the execution of power usage, bandwidth of the network, latency, execution time, and accuracy.

Evaluation of vegetable sauerkraut quality during storage based on convolution neural network

Vegetable sauerkraut is a traditional fermented food. Due to oxidation reactions that occur during storage, the quality and flavor in different periods will change. In this study, the quality evaluation and flavor characteristics of 13 groups of vegetable sauerkraut samples with different storage time were analyzed by using physical and chemical parameters combined with electronic nose. Photographs of samples of various periods were collected, and a convolutional neural network (CNN) framework was established. The relationship between total phenol oxidative decomposition and flavor compounds was linearly negatively correlated. The vegetable sauerkraut during storage can be divided into three categories (full acceptance period, acceptance period and unacceptance period) by principal component analysis and Fisher discriminant analysis. The CNN parameters were fine-tuned based on the classification results, and its output results can reflect the quality changes and flavor characteristics of the samples, and have better fitting, prediction capabilities. After 50 epochs of the model, the accuracy of three sets of data namely training set, validation set and test set recorded 94%, 85% and 93%, respectively. In addition, the accuracy of CNN in identifying different quality sauerkraut was 95.30%. It is proved that the convolutional neural network has excellent performance in predicting the quality of Szechuan Sauerkraut with high reliability.

Atrous convolution aided integrated framework for lung nodule segmentation and classification

Lung cancer has been recognized as the most life-threatening cancer all over the world. Appropriate detection of lung nodule using Computed Tomography (CT) images helps in early stage recognition of lung cancer. Different computer-aided algorithms play an important role in the early diagnosis of lung cancer and can increase the five-year survival rate of lung cancer patients. However, due to structural similarity, manually recognizing the malignant nodule from the benign is time-consuming and challenging task. Recently different deep learning (DL) based Computer-aided diagnosis (CADx) systems have been developed for lung nodule characterization. In this work, an integrated nodule segmentation and characterization framework has been developed using the concept of atrous convolution. The proposed Atrous Convolution-based Convolutional Neural Network (ATCNN) framework can segment and characterize lung nodules by capturing multi-scale features from the HRCT images. Different variants of the ATCNN framework have been analyzed for lung nodule characterization. Among them, ATCNN with a two-layer atrous pyramid and residual connections (ATCNN2PR) has demonstrated the highest classification performance indices for nodule characterization. The new ATCNN2PR framework has obtained an average Dice Similarity Coefficient (DSC), Jaccard Index (JI), and Boundary F1 (BF) score of 0.9715, 0.9520, and 0.9584 for nodule segmentation and sensitivity, specificity, accuracy of 95.84%, 96.89%, and 95.97% for lung nodule characterization on LIDC-IDRI dataset. The proposed automatic trainable end-to-end system has outperforms other competing frameworks by capturing multi-scale features from High-Resolution Computed Tomography (HRCT) nodule images.

A lightweight convolutional neural network for recognition of severity stages of maydis leaf blight disease of maize.

Maydis leaf blight (MLB) of maize (Zea Mays L.), a serious fungal disease, is capable of causing up to 70% damage to the crop under severe conditions. Severity of diseases is considered as one of the important factors for proper crop management and overall crop yield. Therefore, it is quite essential to identify the disease at the earliest possible stage to overcome the yield loss. In this study, we created an image database of maize crop, MDSD (Maydis leaf blight Disease Severity Dataset), containing 1,760 digital images of MLB disease, collected from different agricultural fields and categorized into four groups viz. healthy, low, medium and high severity stages. Next, we proposed a lightweight convolutional neural network (CNN) to identify the severity stages of MLB disease. The proposed network is a simple CNN framework augmented with two modified Inception modules, making it a lightweight and efficient multi-scale feature extractor. The proposed network reported approx. 99.13% classification accuracy with the f1-score of 98.97% on the test images of MDSD. Furthermore, the class-wise accuracy levels were 100% for healthy samples, 98% for low severity samples and 99% for the medium and high severity samples. In addition to that, our network significantly outperforms the popular pretrained models, viz. VGG16, VGG19, InceptionV3, ResNet50, Xception, MobileNetV2, DenseNet121 and NASNetMobile for the MDSD image database. The experimental findings revealed that our proposed lightweight network is excellent in identifying the images of severity stages of MLB disease despite complicated background conditions.

The harmonium model and its unified system view of psychopathology: a validation study by means of a convolutional neural network

The harmonium model (HM) is a recent conceptualization of the unifying view of psychopathology, namely the idea of a general mechanism underpinning all mental disorders (the p factor). According to HM, psychopathology consists of a low dimensional Phase Space of Meaning (PSM), where each dimension of meaning maps a component of the environmental variability. Accordingly, the lower thenumber of independent dimensions in the PSM, and hence its intrinsic complexity, the more limited the way of interpreting the environment. The current simulation study, based on a Convolutional Neural Network (CNN) framework, aims at validating the HM low-dimensionality hypothesis. CNN-based classifiers were employed to simulate normotypical and pathological cognitive processes. Results revealed that normotypical and pathological CNNs were different in terms of both classification performance and layer activation patterns. Using Principal Component Analysis to characterize the PSM associated with the two algorithms, we found that the performance of the normotypical CNN relies on a larger and more evenly distributed number of components, compared with the pathological one. This finding might be indicative of the fact that psychopathology can be modelled as a low-dimensional, poorly modulable PSM, which means the environment is detected through few components of meaning, preventing complex information patterns from being taken into account.

Applying Convolutional Neural Network in Automatic Assessment of Bone Age Using Multi-Stage and Cross-Category Strategy

Bone age is a common indicator of children’s growth. However, traditional bone age assessment methods usually take a long time and are jeopardized by human error. To address the aforementioned problem, we propose an automatic bone age assessment system based on the convolutional neural network (CNN) framework. Generally, bone age assessment is utilized amongst 0–18-year-old children. In order to reduce its variation in terms of regression model building, our system consists of two steps. First, we build a maturity stage classifier to identify the maturity stage, and then build regression models for each maturity stage. In this way, assessing bone age through the use of several independent regression models will reduce the variation and make the assessment of bone age more accurate. Some bone sections are particularly useful for distinguishing certain maturity stages, but may not be effective for other stages, and thus we first perform a rough classification to generally distinguish the maturity stage, and then undertake fine classification. Because the skeleton is constantly growing during bone development, it is not easy to obtain a clear decision boundary between the various stages of maturation. Therefore, we propose a cross-stage class strategy for this problem. In addition, because fewer children undergo X-rays in the early and late stages, this causes an imbalance in the data. Under the cross-stage class strategy, this problem can also be alleviated. In our proposed framework, we utilize an MSCS-CNN (Multi-Step and Cross-Stage CNN). We experiment on our dataset, and the accuracy of the MSCS-CNN in identifying both female and male maturity stages is above 0.96. After determining maturity stage during bone age assessment, we obtain a 0.532 and 0.56 MAE (mean absolute error) for females and males, respectively.

Applicability of multidimensional convolutional neural networks on automated detection of diverse focal liver lesions in multiphase CT images.

To investigate the applicability of multidimensional convolutional neural networks (CNNs) together with multiphase contrast-enhanced CT images on automated detection of diverse focal liver lesions (FLLs). We trained detection models based on 2.5D and 3D CNN frameworks using 567 patients with 3892 FLLs and validated on a relatively large independent cohort of 1436 patients with 4723 lesions. The detection performance across different phases (arterial, portal venous [PV], and combined phases) was assessed for the 2.5D model. The lesions were divided into two groups with a cutoff size of 20mm, and further subdivided into four subgroups of<10, 10-20, 20-50, and ≥50mm, to verify the detection rates for lesions of different sizes for the 2.5D and 3D models. McNemar's test was used to compare the detection sensitivities among different methods. In addition, sensitivity with 95% confidence intervals and free-response receiver operating characteristics (FROC) curves were plotted for visualization of the detectability. In the 2.5D model, the detection rate of PV phase outperformed arterial phase, and a combination of the two phases further improved the performance over a single phase. The detection sensitivities in the arterial, PV, and combined phases were 0.737versus 0.802versus 0.832 for all lesions. The 3D model was superior to the 2.5D model for detecting benign lesions (0.896vs. 0.807, p<0.001), malignant lesions (0.940vs. 0.918, p=0.013), and all lesions (0.902vs. 0.832, p<0.001) regardless of size division. Particularly, the 3D model showed higher sensitivity than the 2.5D model in detecting lesions smaller than 20mm (0.868vs. 0.759, p<0.001). For lesions larger than 20mm, both the 3D and the 2.5D models achieved excellent detection performance. The proposed CNN detection model was demonstrated to adaptively learn the feature representations of diverse FLLs and generalize well to a large-scale validation dataset. The use of multiphase significantly improved the detectability of FLLs compared to single phase. 3D CNN framework showed an enhanced capability over the 2.5D in the detection of FLLs, particularly small lesions. The promising performance shows that the proposed CNN detection system could be a powerful clinical tool for the early detection of hepatic tumors.

Deep learning model for regional solar radiation estimation using satellite images

Solar energy is one of the most prominent renewable energy sources today and it can contribute to the energy market with accurate solar radiation estimation. This study proposes a new deep learning-based solar radiation forecasting model using solar radiation data (W/m2) obtained from a ground-based Meteorology measurement station in Hakkari, Turkey, and satellite images taken simultaneously from the European Meteorological Satellites Utilization Organization (EUMETSAT). The model based on convolutional neural networks (CNN) framework adequately extracts the features inherent in multilayer satellite images used as input and obtains more accurate prediction results than traditional models. The prediction performance of the fully-trained CNN model was evaluated using different error metrics such as mean square error (MSE), mean absolute error (MAE), root mean squared error (RMSE), normalized root mean squared error (nRMSE), and coefficient of determination (R2). The results show that the proposed model can accurately predict solar radiation without the need for any measuring device, by making successful meteorological inferences from satellite images with 0.015% MSE, 8.21% nRMSE and 99.8% R2 ratios. Furthermore, the proposed model can be used to verify and calibrate global solar radiation measured from the ground within other locations around the world with similar geographic and climatological characteristics.

A Filtering Method for LiDAR Point Cloud Based on Multi-Scale CNN with Attention Mechanism

Point cloud filtering is an important prerequisite for three-dimensional surface modeling with high precision based on LiDAR data. To cope with the issues of low filtering accuracy or excessive model complexity in traditional filtering algorithms, this paper proposes a filtering method for LiDAR point cloud based on a multi-scale convolutional neural network incorporated with the attention mechanism. Firstly, a regular image patch centering on each point is constructed based on the elevation information of point clouds. As thus, the point cloud filtering problem is transformed into the image classification problem. Then, considering the ability of multi-scale convolution to extract features at different scales and the potential of the attention mechanism to capture key information in images, a multi-scale convolutional neural network framework is constructed, and the attention mechanism is incorporated to coordinate multi-scale convolution kernel with channel and spatial attention modules. After this, the feature maps of the LiDAR point clouds can be acquired at different scales. For these feature maps, the weights of each channel layer and different spatial regions can be further tuned adaptively, which makes the network training more targeted, thereby improving the model performance for image classification and eventually separating of ground points and non-ground points preferably. Finally, the proposed method is compared with the cloth simulation filtering method (CSF), deep neural network method (DNN), k-nearest neighbor method (KNN), deep convolutional neural network method (DCNN) and scale-irrelevant and terrain-adaptive method (SITA) for the standard ISPRS dataset of point cloud filtering and the filter dataset of Qinghai. The experimental results show that the proposed method can obtain lower classification errors, which proves the superiority of this method in point cloud filtering.

A Fast, Accurate, and Comprehensive PPA Estimation of Convolutional Hardware Accelerators

Convolutional Neural Networks (CNN) are widely adopted for Machine Learning (ML) tasks, such as classification and computer vision. GPUs became the reference platforms for both training and inference phases of CNNs due to their tailored architecture to the CNN operators. However, GPUs are power-hungry architectures. A path to enable the deployment of CNNs in energy-constrained devices is adopting hardware accelerators for the inference phase. However, the literature presents gaps regarding analyses and comparisons of these accelerators to evaluate Power-Performance-Area (PPA) trade-offs. Typically, the literature estimates PPA from the number of executed operations during the inference phase, such as the number of MACs, which may not be a good proxy for PPA. Thus, it is necessary to deliver accurate hardware estimations, enabling design space exploration (DSE) to deploy CNNs according to the design constraints. This work proposes a fast and accurate DSE approach for CNNs using an analytical model fitted from the physical synthesis of hardware accelerators. The model is integrated with CNN frameworks, like TensorFlow, to generate accurate results. The analytic model estimates area, performance, power, energy, and memory accesses. The observed average error comparing the analytical model to the data obtained from the physical synthesis is smaller than 7%.

LPCNN: convolutional neural network for link prediction based on network structured features

In a social network (SN), link prediction (LP) is the process of estimating whether a link will exist in the future. In prior LP papers, heuristics score techniques were used. Recent state-of-the-art studies, like Wesfeiler-Lehman neural machine (WLNM) and learning from subgraphs, embeddings, and attributes for link prediction (SEAL), have demonstrated that heuristics scores may increase LP model accuracy by employing deep learning and sub-graphing techniques. WLNM and SEAL, on the other hand, have some limitations and perform poorly in some kinds of SNs. The goal of this research is to present a new framework for enhancing the effectiveness of LP models throughout various types of social networks while overcoming the constraints of earlier techniques. We present the link prediction based convolutional neural network (LPCNN) framework, which uses deep learning techniques to examine common neighbors and predict relations. Adapts the LP task into an image classification issue and classifies the links using a convolutional neural network. On 10 various types of real-work networks, tested the suggested LP model and compared its performance to heuristics and state-of-the-art approaches. Results revealed that our model outperforms the other LP benchmark approaches with an average area under curved (AUC) above 99%.

Exploring the Potential of Low-Bit Training of Convolutional Neural Networks

Convolutional neural networks (CNNs) have been widely used in many tasks, but training CNNs is time consuming and energy hungry. Using the low-bit integer format has been proved promising for speeding up and improving the energy efficiency of CNN inference, while CNN training can hardly benefit from such a technique because of the following challenges: 1) the integer data format cannot meet the requirements of the data dynamic range in training, resulting in the accuracy drop; 2) the floating-point data format keeps sizeable dynamic range with much more exponent bits, thus using it results in higher accumulation power than using the integer data format; and 3) there are some specially designed data formats (e.g., with group-wise scaling) that have the potential to deal with the former two problems but common hardware platforms cannot support them efficiently. To tackle all these challenges and make the training phase of CNNs benefit from the low-bit format, we propose a low-bit training framework for CNNs to pursue a better tradeoff between accuracy and energy efficiency: 1) we adopt element-wise scaling to increase the dynamic range of data representation, which significantly reduces the quantization error; 2) group-wise scaling with hardware friendly factor format is designed to reduce the element-wise exponent bits without degrading the accuracy; and 3) we design the customized hardware unit that implements the low-bit tensor convolution arithmetic with our multilevel scaling data format. Experiments show that our framework achieves a superior tradeoff between the accuracy and the bit-width than previous low-bit training studies. For training various models on CIFAR-10, using 1-bit mantissa and 2-bit exponent is adequate to keep the accuracy loss within 1%. On larger datasets like ImageNet, using 4-bit mantissa and 2-bit exponent is adequate. Through the energy consumption simulation of the whole network, we can see that training a variety of models with our framework could achieve <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$4.9\times $ </tex-math></inline-formula> – <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$10.2\times $ </tex-math></inline-formula> higher energy efficiency than full-precision arithmetic.

A Multi-Branch Decoder Network Approach to Adaptive Temporal Data Selection and Reconstruction for Big Scientific Simulation Data

A key challenge in scientific simulation is that the simulation outputs often require intensive I/O and storage space to store the results for effective post hoc analysis. This article focuses on a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">quality-aware adaptive temporal data selection and reconstruction</i> problem where the goal is to adaptively select simulation data samples at certain key timesteps in situ and reconstruct the discarded samples with quality assurance during post hoc analysis. This problem is motivated by the limitation of current solutions that a significant amount of simulation data samples are either discarded or aggregated during the sampling process, leading to inaccurate modeling of the simulated phenomena. Two unique challenges exist: 1) the sampling decisions have to be made in situ and adapted to the dynamics of the complex scientific simulation data; 2) the reconstruction error must be strictly bounded to meet the application requirement. To address the above challenges, we develop <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">DeepSample</i> , an error-controlled convolutional neural network framework, that jointly integrates a set of coherent multi-branch deep decoders to effectively reconstruct the simulation data with rigorous quality assurance. The results on two real-world scientific simulation applications show that DeepSample significantly outperforms other state-of-the-art methods on both sampling efficiency and reconstructed simulation data quality.

Master-CAM: Multi-scale fusion guided by Master map for high-quality class activation maps

Class Activation Map (CAM) is one of the most popular approaches to visually explain the convolutional neural networks (CNNs). To obtain fine-grained saliency maps, some works fuse saliency signals of the same image at larger scales. However, existing methods based on multi-scale fusion cannot effectively remove the noise from larger-scale images. In this paper, we propose Master-CAM, which uses Master map to guide multi-scale fusion process to obtain a high-quality class activation map. Master-CAM utilizes the general localization ability of the Master map to reduce the noise of the maps. We call the one with the general localization ability among the saliency maps from the same image as Master map, which is the saliency map of the original-scale input in the multi-scale scenario. In addition, we also present a simple yet effective fusion strategy, Master-Fusion, which is derived from the fusion operation in Master-CAM. Master-Fusion strategy can be easily attached to some saliency methods to improve the performance of these methods. We show through qualitative and quantitative experiments that the proposed Master-CAM outperforms the state-of-the-art methods in different CNN frameworks and datasets.

Correction and pointer reading recognition of circular pointer meter

For the meter images collected in an actual environment, there is the possibility of tilt and rotation. This paper presents a method to calibrate the circular pointer-type meter based on YOLOv5s network. The convolutional neural network framework is used to detect the scale value in the meter panel as the key point. The position information and value information of the detected scale value are used to fit the elliptic equation of the position of the scale value with the least square method for perspective correction and rotation correction of the meter, and the corrected meter image is used to obtain the meter pointer reading. This paper proposes the weighted angle method to read the meter reading. After multiple transformations, the accumulated error of the meter image is eliminated. Finally, comparing the pointer detection method of this paper with the traditional pointer detection method, the error of this detection method is smaller; comparing the meter reading results before and after correction, the meter reading error after correction is 50% less than before correction. Comparing the method in this paper with other mainstream methods, it proves the effectiveness of the our method.

Fish surveys on the move: Adapting automated fish detection and classification frameworks for videos on a remotely operated vehicle in shallow marine waters

Mobile underwater cameras, diver-operated or on underwater vehicles, have become popular for monitoring fisheries. Replacing divers with cameras has clear advantages, such as creating permanent records and accessing waters unavailable to divers. The use of cameras, however, typically produces large quantities of video that are time-consuming to process manually. Automated analysis of underwater videos from stationary cameras using deep learning techniques has advanced considerably in recent years, but the use of mobile cameras potentially raises new challenges for existing methods. We tested how well three automation procedures for stationary underwater cameras, taking an object-centric rather than background-centric approach, performed on surveys of fish using a mobile camera. We analyzed underwear drone videos from reef and seagrass habitat to detect and count two marine fisheries species, luderick (Girella tricuspidata) and yellowfin bream (Acanthopagrus australis). Three convolutional neural network (CNN) frameworks were compared: Detectron Faster R-CNN, Detectron2 Faster R-CNN (using a Regional Proposal Network, RPN), and YOLOv5 (a single-stage detector, SSD). Models performed well overall. Per frame, overall F1 scores ranged 81.4 - 87.3%, precision 88.2 – 96.0%, and recall 73.2 - 88.2%. For quantifying MaxN per video, overall F1 ranged 85.9 – 91.4%, precision 81.9 – 95.3%, and recall 87.1 – 91.1%. For luderick, F1 was &amp;gt; 80% for all frameworks per frame and 89% or higher for MaxN. For yellowfin bream, F1 scores were lower (35.0 - 73.8% for frames, 43.4 - 73.0% for MaxN). Detectron2 performed poorly, and YOLOv5 and Detectron performed similarly with advantages depending on metrics and species. For these two frameworks, performance was as good as in videos from stationary cameras. Our findings show that object detection technology is very useful for extracting fish data from mobile underwater cameras for the system tested here. There is a need now to test performance over a wider range of environments to produce generalizable models. The key steps required area to test and enhance performance: 1. for suites of species in the same habitats with different water clarity, 2. in other coastal environments, 3. trialing cameras moving at different speeds, and 4. using different frame-rates.