Custom Convolutional Neural Network Research Articles

High-Performance Convolutional Neural Network Model to Identify COVID-19 in Medical Images

Convolutional neural network (CNN) is a deep learning (DL) model that has significantly contributed to medical systems because it is very useful in digital image processing. However, CNN has several limitations, such as being prone to overfitting, not being properly trained if there is data duplication, and can cause unwanted results if there is an imbalance in the amount of data in each class. Data augmentation techniques are used to overcome overfitting, eliminate data duplication, and random under sampling methods to balance the amount of data in each class, to overcome these problems. In addition, if the CNN model is not designed properly, the computation is less efficient. Research has proved that data augmentation can prevent or overcome overfitting, eliminating duplicate data can make the model more stable, and balancing the amount of data makes the model unbiased and easy to learn new data as evidenced through model evaluation and testing. The results also show that the custom convolutional neural network model is the best model compared to ResNet50 and VGG19 in terms of accuracy, precision, recall, F1-score, loss performance, and computation time efficiency

Efficient Harris Hawk Optimization (HHO)-Based Framework for Accurate Skin Cancer Prediction

The prediction of skin cancer poses a number of challenges due to the differences in visual characteristics between melanoma, basal cell carcinomas, and squamous cell carcinomas. These visual differences pose difficulties for models in discerning subtle features and patterns accurately. However, a remarkable breakthrough in image analysis using convolutional neural networks (CNNs) has emerged, specifically in the identification of skin cancer from images. Unfortunately, manually designing such neural architectures is prone to errors and consumes substantial time. It has become increasingly popular to design and fine-tune neural networks by using metaheuristic algorithms that are based on natural phenomena. A nature-inspired algorithm is a powerful alternative to traditional algorithms for solving problems, particularly in complex optimization tasks. One such algorithm, the Harris hawk optimization (HHO), has demonstrated promise in automatically identifying the most appropriate solution across a wide range of possibilities, making it suitable for solving complex optimization problems. The purpose of this study is to introduce a novel automated architecture called “HHOForSkin” that combines the power of convolutional neural networks with meta-heuristic optimization techniques. The HHOForSkin framework uses an innovative custom CNN architecture with 26 layers for the analysis of medical images. In addition, a Harris hawk optimization algorithm (HHO) is used to fine-tune the developed model for multiple skin cancer classification problems. The developed model achieves an average accuracy of 99.1% and 98.93% F1 score using a publicly available skin cancer dataset. These results position the developed optimization-based skin cancer detection strategy at the forefront, offering the highest accuracy for seven-class classification problems compared to related works.

Automatic Diagnosis of Coronavirus Using Conditional Generative Adversarial Network (CGAN)

A global pandemic has emerged as a result of the widespread coronavirus disease (COVID-19). Deep learning (DL) techniques are used to diagnose COVID-19 based on many chest X-ray. Due to the scarcity of available X-ray images, the performance of DL for COVID-19 detection is lagging, underdeveloped, and suffering from overfitting. Overfitting happens when a network trains a function with an incredibly high variance to represent the training data perfectly. Consequently, medical images lack the availability of large labeled datasets, and the annotation of medical images is expensive and time-consuming for experts. As the COVID-19 virus is an infectious disease, these datasets are scarce, and it is difficult to get large datasets due to patient privacy. To address these issues by augmenting the COVID-19 dataset. In this paper, we adjusted conditional generation adversarial networks (CGAN) along with traditional augmentation (TA). The augmented dataset includes 6550 X-ray images that can be used to improve the diagnosis of COVID-19, and we have implemented five models of transfer learning procedures (DTL). The proposed procedures yielded high detection accuracy of 95%, 93%, 92%, and 92% in only ten epochs, for VGG-16, VGG-19, Xception, and Inception, respectively, and a custom convolutional neural network. Experimental results prove that our model achieves a high detection accuracy of up to 96% compared to other models. We hope it can be applied in other fields with rare data sets.

An Acoustical and Lexical Machine-Learning Pipeline to Identify Connectional Silences.

Context: Developing scalable methods for conversation analytics is essential for health care communication science and quality improvement. Purpose: To assess the feasibility of automating the identification of a conversational feature, Connectional Silence, which is associated with important patient outcomes. Methods: Using audio recordings from the Palliative Care Communication Research Initiative cohort study, we develop and test an automated measurement pipeline comprising three machine-learning (ML) tools-a random forest algorithm and a custom convolutional neural network that operate in parallel on audio recordings, and subsequently a natural language processing algorithm that uses brief excerpts of automated speech-to-text transcripts. Results: Our ML pipeline identified Connectional Silence with an overall sensitivity of 84% and specificity of 92%. For Emotional and Invitational subtypes, we observed sensitivities of 68% and 67%, and specificities of 95% and 97%, respectively. Conclusion: These findings support the capacity for coordinated and complementary ML methods to fully automate the identification of Connectional Silence in natural hospital-based clinical conversations.

Comparison of Transfer Learning and Established Calibration Transfer Methods for Metal Oxide Semiconductor Gas Sensors

Although metal oxide semiconductors are a promising candidate for accurate indoor air quality assessments, multiple drawbacks of the gas sensors prevent their widespread use. Examples include poor selectivity, instability over time, and sensor poisoning. Complex calibration methods and advanced operation modes can solve some of those drawbacks. However, this leads to long calibration times, which are unsuitable for mass production. In recent years, multiple attempts to solve calibration transfer have been made with the help of direct standardization, orthogonal signal correction, and many more methods. Besides those, a new promising approach is transfer learning from deep learning. This article will compare different calibration transfer methods, including direct standardization, piecewise direct standardization, transfer learning for deep learning models, and global model building. The machine learning methods to calibrate the initial models for calibration transfer are feature extraction, selection, and regression (established methods) and a custom convolutional neural network TCOCNN. It is shown that transfer learning can outperform the other calibration transfer methods regarding the root mean squared error, especially if the initial model is built with multiple sensors. It was possible to reduce the number of calibration samples by up to 99.3% (from 10 days to approximately 2 h) and still achieve an RMSE for acetone of around 18 ppb (15 ppb with extended individual calibration) if six different sensors were used for building the initial model. Furthermore, it was shown that the other calibration transfer methods (direct standardization and piecewise direct standardization) also work reasonably well for both machine learning approaches, primarily when multiple sensors are used for the initial model.

Image classification of degraded polysorbate, protein and silicone oil sub-visible particles detected by flow-imaging microscopy in biopharmaceuticals using a convolutional neural network model

Degradation of polysorbates in biopharmaceutical formulations can induce the formation of sub-visible particles (SvPs) in the form of free-fatty acids (FFAs) and potentially protein aggregates. Flow-imaging microscopy (FIM) is one of the most common techniques for enumerating and characterizing the SvPs, allowing for collection of image data of the SvPs in the size ranges of two to several hundred micrometers. The vast amounts of data obtained with FIM do not allow for rapid manual characterization by an experienced analyst and can be ambiguous. In this work, we present the application of a custom convolutional neural network (CNN) for classification of SvP images of FFAs, proteinaceous particles and silicon oil droplets, by FIM. The network was then used to predict the composition of artificially pooled test samples of unknown and labeled data with varying compositions. Minor misclassifications were observed between the FFAs and proteinaceous particles, considered tolerable for application to pharmaceutical development. The network is considered to be suitable for fast and robust classification of the most common SvPs found during FIM analysis.

Hyperspectral imaging-based detection of soluble solids content of loquat from a small sample

The flavor and market value of loquats are substantially affected by their soluble solids content (SSC). This study explored the ability of hyperspectral imaging combined with multiple regression models to detect SSC of loquats using small sample size. In this study, one hundred and fifteen loquats were collected as samples, and regions of interest in the hyperspectral images of the samples were extracted. The spectral data processed through direct orthogonal signal correction (DOSC), the best among the five processing methods examined, was selected for the feature spectrum process. Successive projection algorithm (SPA), competitive adaptive reweighted sampling (CARS), improved uninformative variable elimination algorithm (imUVE), and imUVE-SPA algorithms were used to extract feature spectra and establish partial least squares regression (PLSR), support vector machine regression (SVR), back-propagation neural network (BPNN), and customized convolutional neural network (CNN) models, respectively. The results demonstrated that the CNN model based on the 18 feature spectra extracted by CARS was the best (correlation coefficient of validation = 0.904, root mean square error of validation = 0.336 %, and residual predictive deviation = 2.339). Consequently, hyperspectral images with small sample size combined with appropriate regression models can be used for nondestructive detection of loquat brix. Furthermore, the wavelength feature extraction algorithm effectively improves the performance of CNN models with small sample sizes. This study can serve as a valuable reference for the nondestructive detection of fruit brix in other limited sample situations.

A Customized ECA-CRNN Model for Emotion Recognition Based on EEG Signals

Electroencephalogram (EEG) signals are electrical signals generated by changes in brain potential. As a significant physiological signal, EEG signals have been applied in various fields, including emotion recognition. However, current deep learning methods based on EEG signals for emotion recognition lack consideration of important aspects and comprehensive analysis of feature extraction interactions. In this paper, we propose a novel model named ECA-CRNN for emotion recognition using EEG signals. Our model integrates the efficient channel attention (ECA-Net) module into our modified combination of a customized convolutional neural network (CNN) and gated circulation unit (GRU), which enables more comprehensive feature extraction, enhances the internal relationship between frequency bands and improves recognition performance. Additionally, we utilize four-dimensional data as input to our model, comprising temporal, spatial and frequency information. The test on the DEAP dataset demonstrates that it enhances the recognition accuracy of EEG signals in both arousal and valence to 95.70% and 95.33%, respectively, while also reducing the standard deviation during five-fold cross-validation to 1.16 and 1.45 for arousal and valence, respectively, surpassing most methods.

A Diabetic Retinopathy Detection Using Customized Convolutional Neural Network

The disease, Diabetic Retinopathy (DR) causes due to damage to retinal blood vessels in diabetic patients. DR occurs if you have type 1 or 2 diabetes along with high blood sugar. When the retinal blood vessels are damaged, they can become clogged, some of which can block the blood supply to the retina leading to blood loss, these new blood vessels may leak, and the creation of scar tissue can lead to loss of vision. It takes a lot of time and effort to examine and analyse fundus images the old-fashioned way to find differences in how the eyes are shaped. In this modern era, technology has evolved so fleet which has the solution to every problem. In this paper, we have proposed a Customized Convolutional Neural Network (CCNN) deep learning technique for Diabetic Retinopathy Detection. We have clung to traditional strategies mainly containing input Data retrieval, pre-processing of data, segmentation, trait measurement, feature extraction, model creation, model training, model testing, consequence, and interpretation of the model. Performance evaluation is done on standard MESSIDOR Dataset in which 560 images for training phase whereas 163 images for testing phase. The experiment results achieved the highest test accuracy of 97.24% which is effectively higher than that of existing algorithms.

Dual Deep CNN for Tumor Brain Classification.

Brain tumor (BT) is a serious issue and potentially deadly disease that receives much attention. However, early detection and identification of tumor type and location are crucial for effective treatment and saving lives. Manual diagnoses are time-consuming and depend on radiologist experts; the increasing number of new cases of brain tumors makes it difficult to process massive and large amounts of data rapidly, as time is a critical factor in patients' lives. Hence, artificial intelligence (AI) is vital for understanding disease and its various types. Several studies proposed different techniques for BT detection and classification. These studies are on machine learning (ML) and deep learning (DL). The ML-based method requires handcrafted or automatic feature extraction algorithms; however, DL becomes superior in self-learning and robust in classification and recognition tasks. This research focuses on classifying three types of tumors using MRI imaging: meningioma, glioma, and pituitary tumors. The proposed DCTN model depends on dual convolutional neural networks with VGG-16 architecture concatenated with custom CNN (convolutional neural networks) architecture. After conducting approximately 22 experiments with different architectures and models, our model reached 100% accuracy during training and 99% during testing. The proposed methodology obtained the highest possible improvement over existing research studies. The solution provides a revolution for healthcare providers that can be used as a different disease classification in the future and save human lives.

CNN-RSVM: a hybrid approach for classification of poikilocytosis using convolutional neural network and radial kernel basis support vector machine

ABSTRACT The most vital blood cells in the body are red blood cells (RBCs), which carry oxygen and nutrients to the tissues. Poikilocytosis refers to a change in the red blood cells’ normal physical shape and structure. The lack of oxygen and nutrients causes defective RBCs to suffer from health problems such as anaemia and stroke. This study proposes a hybrid CNN-SVM model for promptly and precisely classifying poikilocytosis abnormalities. The features of poikilocytosis were extracted using the proposed custom CNN (Convolutional Neural Network) model and classified using the Radial kernel basis SVM (Support Vector Machine) classifier. In comparison to the CNN model with a soft-max classifier and other benchmark work, the hybrid model produced better results. The overall accuracy of the algorithm is 98.21 %, which is higher than the traditional CNN (>5 %) and other benchmark methods. This hybrid CNN-SVM model is promising for the automatic classification of poikilocytosis abnormalities.

A deep learning framework for defect prediction based on thermographic in-situ monitoring in laser powder bed fusion

AbstractThe prediction of porosity is a crucial task for metal based additive manufacturing techniques such as laser powder bed fusion. Short wave infrared thermography as an in-situ monitoring tool enables the measurement of the surface radiosity during the laser exposure. Based on the thermogram data, the thermal history of the component can be reconstructed which is closely related to the resulting mechanical properties and to the formation of porosity in the part. In this study, we present a novel framework for the local prediction of porosity based on extracted features from thermogram data. The framework consists of a data pre-processing workflow and a supervised deep learning classifier architecture. The data pre-processing workflow generates samples from thermogram feature data by including feature information from multiple subsequent layers. Thereby, the prediction of the occurrence of complex process phenomena such as keyhole pores is enabled. A custom convolutional neural network model is used for classification. The model is trained and tested on a dataset from thermographic in-situ monitoring of the manufacturing of an AISI 316L stainless steel test component. The impact of the pre-processing parameters and the local void distribution on the classification performance is studied in detail. The presented model achieves an accuracy of 0.96 and an f1-Score of 0.86 for predicting keyhole porosity in small sub-volumes with a dimension of (700 × 700 × 50) µm3. Furthermore, we show that pre-processing parameters such as the porosity threshold for sample labeling and the number of included subsequent layers are influential for the model performance. Moreover, the model prediction is shown to be sensitive to local porosity changes although it is trained on binary labeled data that disregards the actual sample porosity.

Multi-Class Confidence Detection Using Deep Learning Approach

The advancement of both the fields of Computer Vision (CV) and Artificial Neural Networks (ANNs) has enabled the development of effective automatic systems for analyzing human behavior. It is possible to recognize gestures, which are frequently used by people to communicate information non-verbally, by studying hand movements. So, the main contribution of this research is the collected dataset, which is taken from open-source videos of the relevant subjects that contain actions that depict confidence levels. The dataset contains high-quality frames with minimal bias and less noise. Secondly, we have chosen the domain of confidence determination during social issues such as interviews, discussions, or criminal investigations. Thirdly, the proposed model is a combination of two high-performing models, i.e., CNN (GoogLeNet) and LSTM. GoogLeNet is the state-of-the-art architecture for hand detection and gesture recognition. LSTM prevents the loss of information by keeping temporal data. So the combination of these two outperformed during the training and testing process. This study presents a method to recognize different categories of Self-Efficacy by performing multi-class classification based on the current situation of hand movements using visual data processing and feature extraction. The proposed architecture pre-processes the sequence of images collected from different scenarios, including humans, and their quality frames are extracted. These frames are then processed to extract and analyze the features regarding their body joints and hand position and classify them into four different classes related to efficacy, i.e., confidence, cooperation, confusion, and uncomfortable. The features are extracted using a combination framework of customized Convolutional Neural Network (CNN) layers with Long Short-Term Memory (LSTM) for feature extraction and classification. Remarkable results have been achieved from this study representing 90.48% accuracy with effective recognition of human body gestures through deep learning approaches.

Potato Blight Classification Android Application using Deep learning

Farmers who grow potatoes suffer from significant financial losses each year due several diseases that affect potato plants. The most common diseases are Early and Late Blight , caused by fungus and specific microorganisms, respectively. Early detection and appropriate treatment can save a lot of waste and prevent economic losses. However, traditional visual inspection methods are time-consuming and prone to errors. To address this challenge, we propose a Convolutional Neural Network ( CNN ) approach for plant disease diagnosis. CNNs are a type of deep learning algorithm widely used for image classification tasks. They can automatically learn features from input image data, making them well-suited for plant disease diagnosis. Our customized CNN has fewer trainable parameters, reducing computation time and minimizing information loss. We used several convolutional and pooling layers, followed by fully connected layers with the ReLU activation function. We also applied dropout regularization to prevent overfitting. In conclusion, accurate and efficient plant disease diagnosis is essential for preventing economic losses. Our customized CNN for plant disease diagnosis has the potential to be an effective tool for farmers. It can help them diagnose plant diseases quickly and accurately, leading to timely treatment and reduced loss.

Weld Defect Detection of a CMT Arc-Welded Aluminum Alloy Sheet Based on Arc Sound Signal Processing

The cold metal transfer (CMT) process is widely used in thin plate welding because of its characteristics of low heat input and stable arc. In actual production, a larger weld gap, misalignment, or other problems due to assembly error lead to serious welding defects, such as burn-through and a lack of fusion. The arc sound contains a wealth of information related to the quality of the weld. This work analyzes the mechanism of CMT arc sound generation, as well as the correlation between the time–frequency spectrum of the arc sound signal and welding quality. This paper studies the extraction of the multi-channel time–frequency spectrum of an arc sound and inputs it to a custom convolutional neural network for the CMT welding defect identification of thin aluminum alloy plates. The experimental result shows that the average accuracy of the proposed model is 91.49% in the defect identification of a CMT arc-welded aluminum alloy sheet, which is higher than that of the single-channel time–frequency convolutional neural network and other traditional classification models.

Manuscripts Character Recognition Using Machine Learning and Deep Learning

The automatic character recognition of historic documents gained more attention from scholars recently, due to the big improvements in computer vision, image processing, and digitization. While Neural Networks, the current state-of-the-art models used for image recognition, are very performant, they typically suffer from using large amounts of training data. In our study we manually built our own relatively small dataset of 404 characters by cropping letter images from a popular historic manuscript, the Electronic Beowulf. To compensate for the small dataset we use ImageDataGenerator, a Python library was used to augment our Beowulf manuscript’s dataset. The training dataset was augmented once, twice, and thrice, which we call resampling 1, resampling 2, and resampling 3, respectively. To classify the manuscript’s character images efficiently, we developed a customized Convolutional Neural Network (CNN) model. We conducted a comparative analysis of the results achieved by our proposed model with other machine learning (ML) models such as support vector machine (SVM), K-nearest neighbor (KNN), decision tree (DT), random forest (RF), and XGBoost. We used pretrained models such as VGG16, MobileNet, and ResNet50 to extract features from character images. We then trained and tested the above ML models and recorded the results. Moreover, we validated our proposed CNN model against the well-established MNIST dataset. Our proposed CNN model achieves very good recognition accuracies of 88.67%, 90.91%, and 98.86% in the cases of resampling 1, resampling 2, and resampling 3, respectively, for the Beowulf manuscript’s data. Additionally, our CNN model achieves the benchmark recognition accuracy of 99.03% for the MNIST dataset.

Non-Linear CNN-Based Read Channel for Hard Disk Drive With 30% Error Rate Reduction and Sequential 200-Mbits/s Throughput in 28-nm CMOS

In this work, we present the world first ASIC implementation of a machine learning (ML)-based data detection channel for hard disk drives (HDDs) using a customized convolutional neural network (CNN). The chip demonstrates a 30.3% error rate reduction over the state-of-the-art HDD detection channel with two-dimensional magnetic recording (TDMR) setting. The work incorporates a full-scale co-optimization flow between the ML algorithms and the customized hardware design for achieving: 1) superior detection accuracy; 2) high-throughput sequential detections; and 3) improved power efficiency at the same time. This work also demonstrates for the first time the non-linear detection capability being embedded into the HDD reading channels. The key features of our chip include: 1) a fully unrolled CNN with dedicated silicon and hardware implementation for each convolutional layer to produce fast sequential time-series data detection at 200-Mbits/s; 2) in total six depthwise-separable convolutional layers implemented with two types of systolic arrays and 100% PE utilization for continuous data flow and high pipelining; 3) integer-only convolutions with reduced-precision weights (8 bit) and internal data (6 bit) for detection at an improved detection power efficiency at 0.86 nJ/bit and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\sim $ </tex-math></inline-formula> 4TOPS/W; and 4) customized quantized ReLU (QReLU) units designed to maintain low-precision feature maps and recovery accuracy loss from the model quantization.

Intelligent recognition and integration of grapical helements into virtual surrounding within augmented reality using hybrid convolutional neural networks.

In this paper analysis of modern augmented reality algorithms based on mobile devices was done. As a result, algorithmic shortcomings were identified and the usage of convolutional neural networks was proposed. Within the research the qualitative analysis of modern architectures of convolutional neural networks was carried out and their separate shortcomings at use in systems on the basis of processor architecture ARM was shown. As a result of this research it was found that to achieve the target accuracy and speed of the system it is important to use a hybrid convolutional neural network, which significantly improves the quality criteria of the system. The optimal structure and parameters for initialization and training of a hybrid convolutional neural network system used for augmented reality are obtained. The optimal training sample was formed and the use of pre-trained HCNN on another device of ARM architecture was described.

A deep learning-based framework for retinal fundus image enhancement.

Low-quality fundus images with complex degredation can cause costly re-examinations of patients or inaccurate clinical diagnosis. This study aims to create an automatic fundus macular image enhancement framework to improve low-quality fundus images and remove complex image degradation. We propose a new deep learning-based model that automatically enhances low-quality retinal fundus images that suffer from complex degradation. We collected a dataset, comprising 1068 pairs of high-quality (HQ) and low-quality (LQ) fundus images from the Kangbuk Samsung Hospital's health screening program and ophthalmology department from 2017 to 2019. Then, we used these dataset to develop data augmentation methods to simulate major aspects of retinal image degradation and to propose a customized convolutional neural network (CNN) architecture to enhance LQ images, depending on the nature of the degradation. Peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), r-value (linear index of fuzziness), and proportion of ungradable fundus photographs before and after the enhancement process are calculated to assess the performance of proposed model. A comparative evaluation is conducted on an external database and four different open-source databases. The results of the evaluation on the external test dataset showed an significant increase in PSNR and SSIM compared with the original LQ images. Moreover, PSNR and SSIM increased by over 4 dB and 0.04, respectively compared with the previous state-of-the-art methods (P < 0.05). The proportion of ungradable fundus photographs decreased from 42.6% to 26.4% (P = 0.012). Our enhancement process improves LQ fundus images that suffer from complex degradation significantly. Moreover our customized CNN achieved improved performance over the existing state-of-the-art methods. Overall, our framework can have a clinical impact on reducing re-examinations and improving the accuracy of diagnosis.

Identification of traffic signs for advanced driving assistance systems in smart cities using deep learning.

The ability of Advanced Driving Assistance Systems (ADAS) is to identify and understand all objects around the vehicle under varying driving conditions and environmental factors is critical. Today's vehicles are equipped with advanced driving assistance systems that make driving safer and more comfortable. A camera mounted on the car helps the system recognise and detect traffic signs and alerts the driver about various road conditions, like if construction work is ahead or if speed limits have changed. The goal is to identify the traffic sign and process the image in a minimal processing time. A custom convolutional neural network model is used to classify the traffic signs with higher accuracy than the existing models. Image augmentation techniques are used to expand the dataset artificially, and that allows one to learn how the image looks from different perspectives, such as when viewed from different angles or when it looks blurry due to poor weather conditions. The algorithms used to detect traffic signs are YOLO v3 and YOLO v4-tiny. The proposed solution for detecting a specific set of traffic signs performed well, with an accuracy rate of 95.85%.