Deep Learning Convolutional Neural Network Research Articles

The first use of a photogrammetry drone to estimate population abundance and predict age structure of threatened Sumatran elephants

Wildlife monitoring in tropical rainforests poses additional challenges due to species often being elusive, cryptic, faintly colored, and preferring concealable, or difficult to access habitats. Unmanned aerial vehicles (UAVs) prove promising for wildlife surveys in different ecosystems in tropical forests and can be crucial in conserving inaccessible biodiverse areas and their associated species. Traditional surveys that involve infiltrating animal habitats could adversely affect the habits and behavior of elusive and cryptic species in response to human presence. Moreover, collecting data through traditional surveys to simultaneously estimate the abundance and demographic rates of communities of species is often prohibitively time-intensive and expensive. This study assesses the scope of drones to non-invasively access the Bukit Tigapuluh Landscape (BTL) in Riau-Jambi, Indonesia, and detect individual elephants of interest. A rotary-wing quadcopter with a vision-based sensor was tested to estimate the elephant population size and age structure. We developed hierarchical modeling and deep learning CNN to estimate elephant abundance and age structure. Drones successfully observed 96 distinct individuals at 8 locations out of 11 sampling areas. We obtained an estimate of the elephant population of 151 individuals (95% CI [124, 179]) within the study area and predicted more adult animals than subadults and juvenile individuals in the population. Our calculations may serve as a vital spark for innovation for future UAV survey designs in large areas with complex topographies while reducing operational effort.

Skin cancer classifier based on convolution residual neural network

Accurate automatic classification of skin lesion images is a great challenge as the image features are very close in these images. Convolution neural networks (CNN) promise to provide a potential classifier for skin lesions. This work will present dermatologist-level classification of skin cancer by using residual network (ResNet-50) as a deep learning convolutional neural network (DLCNN) that maps images to class labels. It presents a classifier with a single CNN to automatically recognize benign and malignant skin images. The network inputs are only disease labels and image pixels. About 320 clinical images of the different diseases have been used to train CNN. The model performance has been tested with untrained images from the two labels. This model identifies the most common skin cancers and can be updated with a new unlimited number of images. The DLCNN trained by the ResNet-50 model showed good classification of the benign and malignant skin categories. The ResNet-50 as a DLCNN has verified a significant recognition rate of more than 97% on the testing images, which proves that the benign and malignant lesion skin images are properly classified.

Research on mass image classification based on CNN deep learning

Image classification is one of the most important fundamental problems in computer vision research. It involves many basic problems in the field of computer vision, such as feature extraction, object detection and so on. Therefore, the study of image classification is helpful to understand the basic knowledge in the field of computer vision. Since CNN model made a breakthrough, deep learning has been widely used in image classification field. This paper aims to apply the deep learning method to solve the problem of image classification, study and understand image classification problems to help further study and study in the future. By learning 50000 well classified images, this research tries to find the most suitable algorithm and model to achieve the optimal classification, realizing 20 classification and 100 classifications for 10000 unclassified images with the highest accuracy possible.

Behavioral drowsiness detection system execution based on digital camera and MTCNN deep learning

Drowsy driving is a major cause of road accidents worldwide, necessitating the development of effective drowsiness detection systems. Each year, there are more accidents and fatalities than ever before for a variety of causes. For instance, there were 22,952 fatalities and 79,545 injuries as a result of nearly 66,500 vehicle accidents in the last 10 years. In this paper, we propose a novel approach for detecting drowsiness based on behavioral cues captured by a digital camera and utilizing the multi-task cascaded convolutional neural network (MTCNN) deep learning algorithm. A high-resolution camera records visual indications like closed or open eye movement to base the technique on the driver's behavior. In order to measure a car user's weariness in the present frame of reference, eyes landmarks are evaluated, which results in the identification of a fresh constraint known as "eyes aspect ratio." A picture with a frame rate of 60 frames per second (f/s) and a resolution of 4,320 eyeballs was used. The accuracy of sleepiness detection was more than 99.9% in excellent lighting and higher than 99.8% in poor lighting, according to testing data. The current study did better in terms of sleepiness detection accuracy than a lot of earlier investigations.

Deep transfer learning CNN based for classification quality of organic vegetables

This study introduces a system based on a Convolutional Neural Network (CNN) with deep transfer learning for classifying organic vegetables. It aims to evaluate their quality through artificial intelligence. The approach involves three key steps: collecting data, preparing data, and creating data models. Initially, the data collection phase involves gathering images of organic vegetables from packing facilities, organizing these images into training, testing, and validation datasets. In the preparation phase, image processing techniques are applied to adjust the images for training and testing, resizing each to 224 x 224 pixels. The modeling phase involves using these prepared datasets, which include 3,239 images of two types of organic vegetables, to train the model. The study tests the model's effectiveness using three CNN architectures: Inception V3, VGG16, and ResNet50. It finds that the Inception V3 model achieves the highest accuracy at 85%, VGG16 follows with 82% accuracy, and ResNet50 has the lowest accuracy at 50%. The results suggest that Inception V3 is the most effective at accurately classifying organic vegetables, while VGG16 shows some limitations in certain categories, and ResNet50 is the least effective.

Au-Ag OHCs-based SERS sensor coupled with deep learning CNN algorithm to quantify thiram and pymetrozine in tea.

Pesticide residue detection in food has become increasingly important. Herein, surface-enhanced Raman scattering (SERS) coupled with an intelligent algorithm was developed for the rapid and sensitive detection of pesticide residues in tea. By employing octahedral Cu2O templates, Au-Ag octahedral hollow cages (Au-Ag OHCs) were developed, which improved the surface plasma effect via rough edges and hollow inner structure, amplifying the Raman signals of pesticide molecules. Afterward, convolutional neural network (CNN), partial least squares (PLS), and extreme learning machine (ELM) algorithms were applied for the quantitative prediction of thiram and pymetrozine. CNN algorithms performed optimally for thiram and pymetrozine, with correlation values of 0.995 and 0.977 and detection limits (LOD) of 0.286 and 29ppb, respectively. Accordingly, no significant difference (Pgreater than0.05) was observed between the developed approach and HPLC in detecting tea samples. Hence, the proposed Au-Ag OHCs-based SERS technique could be utilized for quantifying thiram and pymetrozine in tea.

Orchard Mapping with Deep Learning Semantic Segmentation

Abstract: The goal of orchard mapping with deep learning semantic segmentation is to automatic detection and localization of the orchard tree canopy under varied situations like multiple seasons, various tree ages, and varying levels of weed covering. The accuracy of segmentation depends on many factors like season, canopy size, and presence of weeds, background soil condition, and untreated soil. The proposed research have several combinations of training & test data based on orchard conditions The proposed research work contains deep learning convolutional neural network variant. The algorithm U- net network will be used. This helps to automatic detect and localize the tree canopy size using UAV images. Image segmentation will help all farmers by automatically segmenting tree canopies from aerial images for the development of integrated tools. After measurement of these parameters, tree canopy volume can be detected of targeted tree

ECG signal classification based on CNN deep learning

Cardiovascular disease is one of the major health concerns, posing a significant threat to human well-being. Modern diagnostic techniques often rely on electronic devices for assistance. Electrocardiograms (ECG) contain abundant information and serve as a powerful tool in diagnosing cardiovascular diseases. Hence, the vital task of monitoring and categorizing ECG signals is pivotal in averting and identifying cardiovascular ailments. In many hospitals, ECG signal classification often depends on experienced doctors, consuming substantial time and potentially missing the optimal window for disease treatment. This paper presents a methodology utilizing the MIT-BIH arrhythmia database, employing deep neural networks, a 10-layer one-dimensional convolutional neural network, to fulfill the function of ECG signal classification. The model under consideration attains a comprehensive accuracy rate of 99.43% on the test dataset, accompanied by sensitivity (97.86%) and specificity (99.64%). These values surpass other methods mentioned in this paper by 1% to 3%. The experimental findings presented indicate the proposed method holds valuable implications for the future realization of intelligent automated ECG signal classification systems.

Human Activity Recognition Using CNN and Lstm Deep Learning Algorithms

Human Activity Recognition recognizes and classifies the activities performed by the users or people based on the data collected from the sensors of special devices such as smart-watches, smartphones etc. It has become easy to collect a huge amount of data from inertial sensors that are embedded in wearable devices. An accelerometer and gyroscope sensors are most commonly used inertial sensors. There are various already available datasets, in our paper, we are using the Wireless Sensor Data Mining dataset which contains 1,098,207 data of 6 physical activities performed. In this paper, the activities we aim to classify are walking, jogging, going up and downstairs, standing, and sitting. There are various algorithms applied on the various datasets. In our paper, we use Convolutional Neural Network and Long Short-Term Memory deep learning algorithm on the data set, we split the data into training data [80%] and testing data [20%]. By using a confusion matrix, we recognize and classify the activities performed using maximum accuracy.

Enhanced accuracy for classification of video capsule endoscopy images using multiple deep learning convolutional neural networks

Background and aimsVideo capsule endoscopy (VCE) is widely used in the detection of abnormalities in the small intestine. However, there remains the challenge to correctly identify a limited number of possible abnormal images from tens of thousands of total images, and this impediment has limited the expansion of the technology. More recently artificial intelligence (AI) technology has been used in classifying VCE images from patients, but a clinical-grade diagnostic accuracy (greater than 99%) has not been achieved. MethodsThis study proposes a system for the automatic classification of a number of categories of unbounded VCE images with high accuracy by means of a transfer learning approach using multiple convolutional neural networks (CNNs). With this new approach, it is not necessary to implement image segmentation, so that the feature extraction becomes automatic and the existing models can be fine-tuned to obtain specific classifiers. ResultsOver 16,000 VCE gastrointestinal (GI) images from normal individuals including those with normal clean mucosa, the pylorus, the ileocecal valve, reduced mucosal view due to luminal contents and lymphangiectasia (a normal variant), and patients with five pathological states including angioectasia, bleeding, erosion/s, ulcers and foreign bodies, were obtained from a publicly available data set. These were used in building, testing and validating AI models for evaluating the diagnostic accuracy of our combined 17-CNN deep learning approach. Compared to a single CNN approach used by other research groups, our AI method, using 17 CNNs, achieved an overall diagnostic accuracy of 99.79%, with an accuracy of 100% for identifying bleeding and foreign bodies. The high accuracy was further demonstrated in the confusion matrices, precision, recall, and F1 score. ConclusionsWe have developed accurate AI deep learning models for unbounded VCE image classification of various medical conditions in medical practice.

Breast cancer early detection in TP53 SNP protein sequences based on a new Convolutional Neural Network model

INTRODUCTION: Breast cancer (BC) is the most commonly occurring cancer and the second leading cause for women’s disease death. The BC cases are associated with genital mutations which are inherited from older generations or acquired overtime. If the diagnosis is done at the first stage, effects associated with certain treatments can be limited, costs can be saved and the diagnostic time can be minimized. This can also help specialists target the best treatment to increase the rate of cures. Nevertheless, its discovery in patients is very challenging due to silent symptoms aside from the fact the routine screening is not recommended for women under 40 years old.OBJECTIVES: Several efforts are aimed at the BC early detection using machine and deep learning systems. The proposed algorithms use different data types to distinguish between cancerous and non-cancerous cases; as: mammography, ultrasound and MRI (magnetic resonance imaging) images. Then, different learning tools were applied on this data for the classification task. Despite the classification rates which exceed 90%, the major drawback of all these methods is that they are applicable only after the appearance of the cancerous tumors, which reduces the cure rates.METHODS: We propose a new technique for early breast cancer screening. For the data, we focus on cancerous and non-cancerous SNP (Single Nucleotide Polymorphism) protein sequences of the TP53 gene in chromosome 17. This gene is shown to be linked to different single amino acid mutations on which we will shed light here. The method we propose transforms SNP textual sequences into digital vectors via coding. Then, RGB scalogram images are generated using the continuous wavelet transform. A pretreatment of color coefficients is applied to scalograms aiming at creating four different databases. Finally, a CNN deep learning network is used for the binary classification of cancerous and non-cancerous images.RESULTS: During the validation process, we reached good performance with specificity of 97.84%, sensitivity of 96.45%, an overall accuracy of 95.29% and an equal run time of 12 minutes 3 seconds. These values ensure the efficiency of our method.To enhance more these results, we used the ORB feature detection technique. Consequently, the classification rates have been improved to reach 95.9% as accuracyCONCLUSION: Our method will allow significant savings time and lives by detecting the disease in patients whose genetic mutations are beginning to appear.

Optimization of Deep CNN Techniques to Classify Breast Cancer and Predict Relapse

Breast cancer is a fatal disease that has a high rate of morbidity and mortality. Finding the right diagnosis is one of the most crucial steps in breast cancer treatment. Doctors can use machine learning (ML) and deep learning techniques to aid with diagnosis. This work makes an effort to devise a methodology for the classification of Breast cancer into its molecular subtypes and prediction of relapse. The objective is to compare the performance of Deep CNN, Tuned CNN and Hypercomplex-Valued CNN, and infer the results, thus automating the classification process. The traditional method used by doctors to detect is tedious and time consuming. It employs multiple methods, including MRI, CT scanning, aspiration, and blood tests as well as image testing. The proposed approach uses image processing techniques to detect irregular breast tissues in the MRI. The survivors of Breast Cancer are still at risk for relapse after remission, and once the disease relapses, the survival rate is much lower. A thorough analysis of data can potentially identify risk factors and reduce the risk of relapse in the first place. A SVM (Support Vector Machine) module with GridSearchCV for hyperparameter tuning is used to identify patterns in those patients who experience a relapse, so that these patterns can be used to predict the relapse before it occurs. The traditional deep learning CNN model achieved an accuracy of 27%, the tuned CNN model achieved an accuracy of 92% and the hypercomplex-valued CNN achieved an accuracy of 98%. The SVM model achieved an accuracy of 89% and on tuning the hyperparameters by using GridSearchCV it achieved and accuracy of 98%.

Utilizing Deep Learning Algorithms for Signal Processing in Electrochemical Biosensors: From Data Augmentation to Detection and Quantification of Chemicals of Interest.

Nanomaterial-based aptasensors serve as useful instruments for detecting small biological entities. This work utilizes data gathered from three electrochemical aptamer-based sensors varying in receptors, analytes of interest, and lengths of signals. Our ultimate objective was the automatic detection and quantification of target analytes from a segment of the signal recorded by these sensors. Initially, we proposed a data augmentation method using conditional variational autoencoders to address data scarcity. Secondly, we employed recurrent-based networks for signal extrapolation, ensuring uniform signal lengths. In the third step, we developed seven deep learning classification models (GRU, unidirectional LSTM (ULSTM), bidirectional LSTM (BLSTM), ConvGRU, ConvULSTM, ConvBLSTM, and CNN) to identify and quantify specific analyte concentrations for six distinct classes, ranging from the absence of analyte to 10 μM. Finally, the second classification model was created to distinguish between abnormal and normal data segments, detect the presence or absence of analytes in the sample, and, if detected, identify the specific analyte and quantify its concentration. Evaluating the time series forecasting showed that the GRU-based network outperformed two other ULSTM and BLSTM networks. Regarding classification models, it turned out signal extrapolation was not effective in improving the classification performance. Comparing the role of the network architectures in classification performance, the result showed that hybrid networks, including both convolutional and recurrent layers and CNN networks, achieved 82% to 99% accuracy across all three datasets. Utilizing short-term Fourier transform (STFT) as the preprocessing technique improved the performance of all datasets with accuracies from 84% to 99%. These findings underscore the effectiveness of suitable data preprocessing methods in enhancing neural network performance, enabling automatic analyte identification and quantification from electrochemical aptasensor signals.

A convolutional neural network deep learning method for model class selection

AbstractThe response‐only model class selection capability of a novel deep convolutional neural network method is examined herein in a simple, yet effective, manner. Specifically, the responses from a unique degree of freedom along with their class information train and validate a one‐dimensional convolutional neural network. In doing so, the network selects the model class of new and unlabeled signals without the need of the system input information, or full system identification. An optional physics‐based algorithm enhancement is also examined using the Kalman filter to fuse the system response signals using the kinematics constraints of the acceleration and displacement data. Importantly, the method is shown to select the model class in slight signal variations attributed to the damping behavior or hysteresis behavior on both linear and nonlinear dynamic systems, as well as on a 3D building finite element model, providing a powerful tool for structural health monitoring applications.

Optimization of Briquette Classification Using Deep Learning

Indonesia needs energy to meet its needs. Energy sourced from waste is called biomass briquettes. The manufacture of briquettes is still done in a traditional way, so there are product quality problems. A computational approach has been taken to base the quality of the product on certain characteristics so that the types of briquettes that are made can be classified or sorted. The research objective of this work is to determine the quality of briquettes. The approach uses deep learning methods and convolutional neural network (CNN) architecture. Classification is based on good and bad briquette products, and testing is based on the level of performance accuracy. The dataset formed consists of 5,280 images. As training data, 85% of data is used and 15% is used as test data using 300 epoch parameters, 32 batch sizes, and learning speed up to 0.001. The results of testing and evaluating the performance of briquette products with the general CNN architecture have a level of accuracy that is not optimal. The optimization results of testing the MobileNetV2 architecture with a ratio of 70:30 obtains predictions with 0.99 as the highest accuracy value and 0.73 as the lowest. The average predicted value of the model is 0.95. The computational approach is able to provide traditional communities with solutions for the process of making briquettes.

A novel hybrid model in the diagnosis and classification of Alzheimer's disease using EEG signals: Deep ensemble learning (DEL) approach

Recent years have witnessed a surge of sophisticated computer-aided diagnosis techniques involving Artificial Intelligence (AI) to accurately diagnose and classify Alzheimer’s disease (AD) and other forms of Dementia. Despite these advancements, there is still a lack of reliable and accurate methods for distinguishing between (AD) and Healthy Controls (HC) using Electroencephalography signals (EEG). The main challenge is finding the right features from the intricate spectral-temporal EEG data, which can provide information sufficient for diagnosis. This study proposes a new approach integrating Deep Ensemble Learning (DEL) and 2-dimensional Convolutional Neural Networks (2D-CNN) to address these issues. Combining state-of-the-art supervised deep learning algorithms within an ensemble model architecture aims to accurately diagnose and classify EEG signals of AD and HC subjects. Public EEG-based Alzheimer's datasets have been classified in the DEL model without applying any feature extraction after cleaning from noise and artifacts.Furthermore, the proposed DEL model used 5 different 2D-CNN models as internal classifiers. As a result, the EEG-based DEL model proposed for the first time provided high accuracy in AD classification. The proposed DEL model reached an average accuracy of 97.9% in AD classification due to 5 cross-fold training.In conclusion, this work renders that incorporating ensemble learning techniques into automotive health applications create extensible and stable AI models needed for computer-aided diagnostic. However, although the reported results and evaluation are promising,further efforts will need to be made to improve the accuracy of our proposed model. In addition, a fine-grid evaluation will be necessary to accurately understand potential impacts in clinical applications, such as earlier diagnosis or treatment decisions.

Assessing the impact of borehole field data on AI-based deep learning models for heating and cooling prediction

Sensor readings play a critical role in the prediction performance of AI-based data-driven fault-detection and diagnostics for borehole heat exchangers in heating and cooling systems. The accuracy and algorithm selection of individual deep neural networks are well-investigated research topics but the impact of quality of sensors data on algorithm selection and its further impact on prediction performance has not been investigated and quantified for geothermal heating/cooling systems. The objective of the current work is to investigate the impact of data quality, available as measured time-series data from the borehole field, on the prediction accuracy of the deep neural networks-based models. In the current work four different data sets (D-1, D-2, D-3, D-4) are used for data-driven modeling using four different deep machine learning approaches, i.e., LSTM, BD-LSTM, GRU and CNN. The developed AI models have been trained, validated, and tested using the real-field data sets with same number of independent input parameters i.e., inlet temperature and mass flow rate to predict the fluid outlet temperature. The validated AI models are used to predict the performance of borehole heat exchanger to meet the heating and cooling loads. The prediction results show that the BD-LSTM outperforms other deep neural networks in terms of accuracy. With the data sets D-1, D-2, D-3 and D-4 it was observed that BD-LSTM had lowest MAPE of 2.99 %, 0.22 %, 6.39 % and 2.21 % respectively. Several experiments show that the BD-LSTM and LSTM models are better than the other used approaches in terms of prediction accuracy, but larger number of trainable parameters results in more computation time than CNN and GRU respectively. More data points with higher sampling frequency, results in more accurate prediction as depicted by the results from D-2 and D-4. Although, the results showed that the data collection interval is crucial, but different test runs indicate that avoiding the loss of time-series data during the cleaning process is equally important to accurately capture the Seasonality and Trend of the heating/cooling data. This is realized in case of D-1 where almost 58 % of the values were removed as outliers and these missing values were needed to be approximated which also affected the model prediction accuracy. The presented analysis highlights that the LSTM based DNN model has the capability to predict borehole heat exchanger performance accurately and the predictions are highly dependent on data frequency, the quality as well as data quantity from geothermal installations used to train the model.

Block-based compressive sensing in deep learning using AlexNet for vegetable classification

Vegetables can be distinguished according to differences in color, shape, and texture. The deep learning convolutional neural network (CNN) method is a technique that can be used to classify types of vegetables for various applications in agriculture. This study proposes a vegetable classification technique that uses the CNN AlexNet model and applies compressive sensing (CS) to reduce computing time and save storage space. In CS, discrete cosine transform (DCT) is applied for the sparsing process, Gaussian distribution for sampling, and orthogonal matching pursuit (OMP) for reconstruction. Simulation results on 600 images for four types of vegetables showed a maximum test accuracy of 98% for the AlexNet method, while the combined block-based CS using the AlexNet method produced a maximum accuracy of 96.66% with a compression ratio of 2×. Our results indicated that AlexNet CNN architecture and block-based CS in AlexNet can classify vegetable images better than previous methods.

Cross-device target migration intelligent diagnosis method of wind power gearbox faults

The current convolutional neural network (CNN) deep learning algorithm for wind power gearbox fault diagnosis requires a substantial amount of high-quality tagged data from testbeds to achieve accurate results. However, it may not perform well in practical engineering scenarios where there is limited fault tag data and incomplete fault information. To overcome this challenge, a novel approach utilizing cross-device target transfer for intelligent diagnosis of wind turbine gearbox faults is proposed. This method transforms the fault diagnosis of wind turbine gearboxes into an image recognition and classification task. Initially, the source domain gearbox testbed data is subjected to dimension transformation using the Gramian angular difference field technique, resulting in the generation of small sample two-dimensional images. This transformation simulates the limited fault tag data scenario encountered in practical engineering. Once the fault features have been augmented, the small sample source domain gearbox testbed data is fed into the IConvNeXt deep CNNs. These CNNs are interconnected and specifically designed for the field. They aim to extract and recognize fault features, thereby creating a pre-training model. The pre-training model is then transferred to the target domain wind power gearbox data for fault diagnosis experiments. The results of these experiments demonstrate the significant effectiveness of the proposed method in recognizing faults in small sample wind power gearboxes within the target domain. Overall, this method successfully achieves the goal of intelligent fault diagnosis through target transfer. It addresses the challenge of limited fault tag data and incomplete fault information, thereby enhancing the performance of CNNs in practical engineering scenarios.

Identifying stable EEG patterns over time for mental workload recognition using transfer DS-CNN framework

Timely evaluating operator Mental Workload (MW) levels in human–computer interaction helps to assign tasks legitimately and guarantee safety and efficiency at the same time. This study presented a stable Electroencephalogram (EEG) pattern over time. A data augmentation (DA) method and a deep neural network (NN) are developed accordingly based on the peculiarity of the EEG pattern for MW assessment. The EEG pattern consists of rhythmic energies in the Temporal–Spatial–Spectral domain, named energy tensor. The DA method for energy tensors is realized by randomly disrupting the temporal dimension which was certified to be irrelevant to MW levels in our previous study. The NN is designed based on deep transfer learning and Depthwise Separable Convolutional Neural Networks (DS-CNN). Energy tensors after DA are used as the input of the proposed NN to construct MW classifiers and the entire operation is named transfer DS-CNN framework. The experiment results demonstrate that the optimal temporal resolution of EEG patterns for MW classification is 3.5 min. The transfer DS-CNN framework is effective enough to build a cross-session MW classifier for a specific subject in an average of 9.195 s, with an average prediction accuracy of 97.22%. Our study suggests that the stable EEG patterns and the DA method preserve features for cross-session MW recognition. It is befitting to discriminate MW levels in real-time with the transfer DS-CNN framework.