Search

The classification of underwater acoustic targets is a challenging task due to the intricacies involved in soundscape, huge amounts of background interference, and the varied features of underwater targets. Various approaches have been explored for classifying underwater targets based on acoustic signatures. The problem with these conventional methods lies in the extensive domain-specific knowledge they demand for effective feature engineering. Alternatively, the deep learning approach is of great appeal for assistance to manual sonar operators who mainly rely on their expertise for the classification of targets. This work leverages the approach of deep neural networks for underwater acoustic target categorization, aiming to analyze whether we can reduce the need for extensive domain knowledge required, by letting deep learning algorithms itself find the deep audio embeddings. This approach implements an audio classifier designed to classify acoustic targets with input as log Mel-spectrograms. To compare the performance of this classifier, we also implement a regular neural network for audio classification with MFCCs as input features. The overall results of the model are promising. [ABSTRACT FROM AUTHOR]

The image-based automatic plant disease identification has emerged as a prominent research area, driven by recent advancements in machine learning algorithms. Conventional disease detection methods rely on manual feature extraction from acquired images, involving extensive expertise from experts and pathologists. This labor-intensive practice is often time-consuming and impractical. With the recent advancements in digital technologies and computer vision, Deep neural networks address these challenges by learning the features automatically. This paper focuses on the EfficientNet B0 architecture-based deep neural network model integrated with a convolutional block attention mechanism module for identifying major wheat crop diseases. The attention module in the EfficientNet B0 model improves its performance by emphasizing the disease-infected regions in leaf images. To evaluate the proposed methodology, experiments were conducted on the WheatRust21 image dataset, comprising 6556 images of three major fungal diseases, namely, stripe rust, leaf rust, and stem rust, collected in natural field conditions from 2019 to 2021. Our developed EfficientNet B0-CBAM model performed well with a testing accuracy of 98.70% and an impressive F1 score of 98% on the dataset. The Grad-CAM visualization demonstrates that the attention module focuses more on the infected region and plays a crucial role in wheat disease identification. The results are quantitatively substantial, with the model achieving high accuracy and other evaluation metrics scores with an average of 98.5 %. These findings hold promise for practical applications in agriculture and sustainable crop management. [ABSTRACT FROM AUTHOR]

[Display omitted] • Three types of PD signals (corona, internal, and surface) are collected experimentally and denoised using Symlet4 before extracting input features for classification study. • ML approaches incorporate statistical characteristics, while the CNN model emphasizes CWT-based scalogram images. • Enhancing the PD classification with the help of CWT-based scalogram images using customize CNN model with an appropriate hyperparameter setting. • Classification performance is also analyzed using ML models (ANN, SVM, KNN, and RF) with the proposed CNN model. The purpose of this work is to classify different types of partial discharge (PD) signals including corona, surface, and internal discharges. The collected PD signals are de-noised using symlet4 mother wavelet and then subjected to classification using machine learning (ML) and convolutional neural network (CNN) architecture. Statistical features (skewness, kurtosis, standard deviation, variance, etc.) are extracted which are used to train the ML models, whereas the CNN model is emphasized automated feature extraction from a novel approach based on scalogram images. From the denoised signals, continuous wavelet transforms (CWT) based scalogram images are obtained which highlight the resolution of the signal's energy and are used for feature extraction in the classification study. In this work, PD signals are used for both ML and CNN model training, validation, and testing. Classification results demonstrate that the proposed CNN model achieves a recognition accuracy of up to 100% compared with support vector machines (SVM), artificial neural network (ANN), k- nearest neighbors (KNN), and random forest (RF) based ML classifiers which are 94.26%, 94.26%, 91%, and 95.1% respectively. [ABSTRACT FROM AUTHOR]

Nowadays, Human Activity Recognition (HAR) is a key research area with many ubiquitous innovative solutions, where both accelerometer and gyroscope data provide information about an observed person's physical activity. HAR offers a diverse variety of important applications, including healthcare, burglary detection, workplace monitoring, and emergency identification. Traditional recognition approaches rely on extracting handmade features from the obtained data to identify the type of human action. Additionally, the efficacy of these works is dependent upon the specific customized features that are chosen. One potential approach to tackle this issue is to utilize Convolutional Neural Networks (CNN) to automatically learn the relevant features. In this paper, we propose a deep learning model, WISNet, a custom 1D-CNN approach to recognize six complex human activities: Jogging, Walking Downstairs, Sitting, Standing, Walking and Climbing Upstairs. The model includes a Convolved Normalized Pooled (CNP M) Block to generate significant features from the initial layers. An Identity and Basic (IDB N) Block is incorporated to extract residual progressive features for capturing complex sequential data dependencies. Channel and Spatial attention (CAS b) Block is integrated with the network to prioritize or minimize essential features based on relative weights. The proposed WISNet model achieved an enhanced accuracy and F1-score of 96.41 % and 0.95 for the HAR dataset by surpassing the existing transfer learning architectures such as Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), and Recurrent Neural Network (SimpleRNN). By strategically integrating the CNP M , IDB N , and CAS b blocks, this study aims to tackle distinct challenges encountered in the classification process by enhancing the discernment of features essential for the precise identification of multi-class human activity recognition. The seamless integration of these blocks within the model plays a pivotal role in elevating the overall performance of the WISNet architecture. The work also validates WISNet with similar open-source datasets (UCI-HAR and KU-HAR) and dissimilar open-source datasets (Sleep state detection, Fall detection, and ECG Heartbeat). [ABSTRACT FROM AUTHOR]

The implementation of deep learning algorithms has contributed to various applications related to the detection of fruit quality. The quality attributes of fruit such as total soluble solids, moisture content, pH, colour changes, and firmness at different varieties can be predicted according to different storage conditions with reliable classification accuracy. The advances in non-destructive techniques have led to the rapid utilisation of the imaging approach in order to monitor the fruit quality. These image datasets encompass diverse information which requires extensive data extraction. To overcome this issue, a deep learning approach using convolutional neural network was used to evaluate the fruit quality. A graphical user interface-based software (DLFRUIT-GUI) for data processing of fruit quality is developed. The toolbox allows the model training and selection based on the image datasets of the fruit. The software offers a push-button approach to establish deep learning models for monitoring fruit quality. The adoption of convolutional neural network model successfully improves the model performance which demonstrated efficient results in predicting the fruit quality at different varieties according to various storage conditions. The DLFRUIT-GUI toolbox provides rapid monitoring of fruit quality detection that can easily be accessible by users who have no programming skills and tedious data analysis. • A software development is presented to predict fruit quality. • Fruit quality at various storage conditions with reliable classification accuracy. • Deep learning approach using convolutional neural network is investigated. • DLFRUIT-GUI provides rapid monitoring of fruit quality detection. [ABSTRACT FROM AUTHOR]

Estimation of aquatic ecosystem health indices can assist in reducing the burden of time-consuming, labor-intensive, and cost-effective fieldwork for the sustainable evaluation of freshwater ecosystem status. In this study, we developed a deep neural network to estimate the trophic diatom index (TDI), benthic macroinvertebrate index (BMI), and fish assessment index (FAI) using water quality and hydraulic and hydrological data. A convolutional neural network (CNN) model was built to estimate health indices. In addition, an autoencoder was adopted to produce manifold features that were used as inputs for the CNN model. Conventional machine learning models, including artificial neural networks, support vector machines, random forests, and extreme gradient boosting, have been developed to estimate the TDI, BMI, and FAI. The results showed that the CNN with an autoencoder exhibited the best performance, with validation accuracies of Nash Sutcliffe Efficiency (NSE) and root mean squared error (RMSE) values of 0.592 and 17.249 for TDI, 0.669 and 12.282 for BMI, and 0.638 and 13.897 for FAI, respectively. The autoencoder enhanced the nonlinear feature learning of the time series and static input data, which contributed to improving the CNN feature extraction for accurate estimation of aquatic ecosystem health indices compared to other data-driven approaches. Therefore, deep learning techniques can be used to investigate aquatic ecosystem health by successfully reflecting the quantitative and qualitative features of health indices. [Display omitted] • CNN linked with autoencoder directly evaluated aquatic ecosystem health indices. • Autoencoder conducted nonlinear data mapping of 81 static and time-series feature. • CNN with autoencoder improved estimation accuracy of aquatic ecosystem health indices. • Fusion of deep learning models showed a potential for analyzing aquatic ecosystem. [ABSTRACT FROM AUTHOR]

The effective detection of synthetic antioxidants by surface-enhanced Raman scattering (SERS) is still challenging due to the presence of phenolic hydroxyl groups in their molecular structures which is difficult to be adsorbed on the conventional SERS substrates, resulting in weak Raman scattering signals. A diazo derivatization reaction was introduced to convert antioxidant molecules into azo derivatives through a diazo derivatization reaction, enabling the amplification of weak Raman signals via the N=N double bond with strong vibrational modes. And excellent linearity was achieved for the antioxidants by a deep learning algorithm called the convolutional neural networks regression model to automatically extract features from the entire spectrum and reduce the signal fluctuation mitigation. [Display omitted] • Diazo derivatization boosts Raman signals of antioxidants by inducing strong N=N vibrations. • A CNN regression model was applied to facilitate the automatic extraction of features from the spectrum and obtain the enlarged linear range. • The SERS method with diazo derivatization detects antioxidants quickly and sensitively in 20 mins. Synthetic antioxidants serve as essential protectors against oxidation and deterioration of edible oils, however, prudent evaluation is necessary regarding potential health risks associated with excessive intake. The direct adsorption of antioxidants onto conventional surface-enhanced Raman scattering (SERS) substrates is challenging due to the presence of phenolic hydroxyl groups in their molecular structures, resulting in weak Raman scattering signals and rendering direct SERS detection difficult. In this study, a diazo derivatization reaction was employed to enhance SERS signals by converting antioxidant molecules into azo derivatives, enabling the amplification of the weak Raman scattering signals through the strong vibrational modes induced by the N = N double bond. The resulting diazo derivatives were characterized using UV–visible absorption and infrared spectroscopy, confirming the occurrence of diazo derivatization of the antioxidants. The proposed method successfully achieved the rapid detection of three commonly used synthetic antioxidants, namely butylated hydroxyanisole (BHA), tert -butylhydroquinone (TBHQ), and propyl gallate (PG) on interfacial self-assembled gold nanoparticles. Furthermore, rapid predictions of BHA, PG, and TBHQ within the concentration range of 1 × 10−6 to 2 × 10−3 mol/L were achieved by integrating a convolutional neural network model. The predictive range of this model surpassed the traditional quantitative method of manually selecting characteristic peaks, with linear coefficients (R2) of 0.9992, 0.9997, and 0.9997, respectively. The recovery of antioxidants in real soybean oil samples ranged from 73.0 % to 126.4 %. Based on diazo derivatization, the proposed SERS method eliminates the need for complex substrates and enables the analysis and determination of synthetic antioxidants in edible oils within 20 min, providing a convenient analytical approach for quality control in the food industry. [ABSTRACT FROM AUTHOR]

The human respiratory system might be seriously affected by COVID-19 infection. Therefore, early classification of it is a crucial task. Quantum machine learning and quantum neural network models can play an effective role in multiclass classification problems. Compared to standard deep and machine learning classifiers, the quantum variational classifier may lead to less memory usage, accuracy, and portability for respiratory disease detection. This article proposes a hybrid respiratory lung disease detection framework based on classical CNN and Quantum classifiers. It combines a classical deep feature extraction model with quantum classifiers. A new custom convolutional neural network (CCNN) deep learning model is proposed to perform feature extraction, and the Multi-Multi-Single (MMS) & Multi-Single-Multi-Single (MSMS) are proposed as quantum machine learning algorithms. These two quantum classifiers are built via a quantum variational circuit with encoding, entanglement, and measurement properties. The tests were carried out on the COVID-19 Radiography Dataset (CRD), which contains 15,153 chest X-ray images of COVID-19, Viral, and Normal. The experimental results revealed that the proposed model had the highest training and testing accuracy of 98.9% and 98.1%, on the CRD dataset, with a computation cost of 0.07 and 0.08 respectively. This hybrid model performs better than the other standard deep learning models. Additionally, we validated our MMS and MSMS quantum classifiers by deploying them on the IBM Q-QASM real-time quantum computer. • A New Hybrid Custom CNN (CCNN) with a Quantum-Classical Network is proposed using lung chest X-ray images to identify respiratory disorders. • Encoding qubits, producing entangled quantum states, monitoring classical computer output, and expanding qubits optimise classical trainable parameters. • To demonstrate their efficacy, we compared MMS and MSMS to various state-of-art models like Mobile Net, VGG16, Inception Net, and ResNet50. • An actual quantum processor, IBMQ-QASM, tests the improved custom CNN Quantum-classical model, which is proven to handle the noisy constraints of retrieved features in images with a high resolution. • The experiment outperformed earlier models in qubits, qdepth, training, and testing accuracies, concerning various evaluation metrics. [ABSTRACT FROM AUTHOR]

• Propose a new approach to conduct field tests and modeling works together. • Compare three physics-based approaches and three machine-learning-based methods. • Compare steady-state and transient simulations using the six different approaches. • Analyze the delay problem in the transient simulation using deep learning methods. Conventionally, researchers conducted field tests and modeling works for the Variable Refrigerant Flow system separately. In this study, we used the compressor model as a case study to illustrate a novel approach to integrate field tests and modeling works. Field tests in an office building were conducted to collect data. For mass flow rate prediction, three traditional models, including the 20-coefficient model, the efficiency-based model, and the efficiency-based 20-coefficient model, and three machine-learning-based models including Support Vector Regression, Neural Network, and Random Forest, were investigated and compared. We found that the efficiency-based 20-coefficient model and the Support Vector Regression model had a higher accuracy (the mean relative errors were 0.11% and 0.15%, and the coefficient of variation of the root mean-square-error were 0.20 and 0.23) and lower uncertainty within 0.2 g•s−1 for steady-state operation prediction. However, all six models failed to predict accurately the beginning part of the transient process. A tens-of-seconds delay existed between the predicted values and the experiment values. We applied the Convolutional-Neural-Network-based model to address this problem. The mean relative error of this model is reduced to 2% for dynamic simulation. In summary, we recommend the efficiency-based 20-coefficient model, and the Support Vector Regression model for the steady-state compressor model development, while the Convolutional-Neural-Network-based model is recommended for transient model development. For the power consumption prediction, 20-coefficient and Neural Network models can predict transient data well. The process of predicting the capacity is the same as the process of mass flow rate prediction. [ABSTRACT FROM AUTHOR]

• Conventional quality analysis of emulsions proved subjective and time-consuming. • Machine learning outperformed deep learning predicting in-process emulsion quality. • Droplet characteristics are better predictors compared to pixel intensity. • Random Forest model presented 99.8% accuracy. • The automated approach demonstrated promising potential in reducing over processing. Determining the uniformity and consistency of droplet size in the dispersed phase is key to emulsion stability. Conventional methods, which include manual microscopic evaluation and laser diffraction, have presented many challenges to achieve an accurate evaluation of droplet dispersion in pharmaceutical emulsions. Artificial intelligence techniques have demonstrated potential in overcoming the subjectivity and time-consumption associated with the manual approaches in industry. A new automated machine learning approach is presented in this study to predict in-process emulsion quality from micrographs. Droplet characteristics are extracted from emulsion micrographs using a histogram-based image segmentation technique. Machine learning classification models are developed, with a selected set of droplet characteristics as predictors, via Random Forest, Multinomial Logistic Regression and Vanilla Neural Network to classify the micrographs into four categories. The hyper-parameters of the models are tuned using 10-fold cross validation. A pixel-based Convolutional Neural Network model is also investigated. Random Forest presented the best accuracy of 99.78% compared to the deep learning models, which presented a bias towards the high frequency classes. The automated machine learning approach demonstrated promising potential for inline emulsion quality evaluation. [ABSTRACT FROM AUTHOR]

Surface monitoring, vertical atmospheric column observation, and simulation using chemical transportation models are three dominant approaches for perception of fine particles with diameters less than 2.5 micrometers (PM 2.5) concentration. Here we explored an image-based methodology with a deep learning approach and machine learning approach to extend the ability on PM 2.5 perception. Using 6976 images combined with daily weather conditions and hourly time data in Shanghai (2016), trained by hourly surface monitoring concentrations, an end-to-end model consisting of convolutional neural network and gradient boosting machine (GBM) was constructed. The mean absolute error, the root-mean-square error and the R -squared for PM 2.5 concentration estimation using our proposed method is 3.56, 10.02, and 0.85 respectively. The transferability analysis showed that networks trained in Shanghai, fine-tuned with only 10% of images in other locations, achieved performances similar to ones from trained on data from target locations themselves. The sensitivity of different regions in the image to PM 2.5 concentration was also quantified through the analysis of feature importance in GBM. All the required inputs in this study are commonly available, which greatly improved the accessibility of PM 2.5 concentration for placed and period with no surface observation. And this study makes an exploratory attempt on pollution monitoring using graph theory and deep learning approach. Image, graphical abstract [ABSTRACT FROM AUTHOR]

• A model to detect distraction of drivers. • Utilizes Convolutional Neural Networks to detect the activity being performed by the driver. • The model is able to differentiate between the types of distractions as well. • High detection accuracy of 99% achieved over a large dataset. • The model proposed, takes significantly less training time and delivers high test accuracy. With the intervention of social media and internet technology, people are getting more and more careless and distracted while driving which is having a severe detrimental effect on the safety of the driver and his fellow passengers. To provide an effective solution, this paper puts forward a Machine Learning model using Convolutional Neural Networks to not only detect the distracted driver but also identify the cause of his distraction by analyzing the images obtained using the camera module installed inside the vehicle. Convolutional Neural Networks are known to learn spatial features from images, which can be further examined by fully connected neural networks. The experimental results show a 99% average accuracy in distraction recognition and hence strongly support that our Convolutional Neural Networks model can be used to identify distraction among the drivers. [ABSTRACT FROM AUTHOR]

In order to make financial investment more stable and more profitable, convolutional neural network (CNN) and deep learning technology are used to quantify financial investment, so as to obtain more robust investment and returns. With the continuous development of in-depth learning technology, people are applying it more and more widely. Deep learning is put forward on the basis of neural network. It contains more hidden layers, shows more powerful learning ability, and can abstract data at a higher level, so as to obtain more accurate data. CNN is a multi-layer network structure which simulates the operation mechanism of biological vision system. Its special structure can obtain more useful feature descriptions from original data and is very effective in extracting data. Therefore, in this study, the two technologies are combined to quantify financial investment. The results show that the convolution neural network and deep learning algorithm can obtain relatively accurate investment strategies, thus ensuring investment returns and reducing investment risks. [ABSTRACT FROM AUTHOR]

Artificial intelligence (AI) is increasing its role in diagnosis of patients with suspicious coronary artery disease. The aim of this manuscript is to develop a deep convolutional neural network (CNN) to classify coronary computed tomography angiography (CCTA) in the correct Coronary Artery Disease Reporting and Data System (CAD-RADS) category. Two hundred eighty eight patients who underwent clinically indicated CCTA were included in this single-center retrospective study. The CCTAs were stratified by CAD-RADS scores by expert readers and considered as reference standard. A deep CNN was designed and tested on the CCTA dataset and compared to on-site reading. The deep CNN analyzed the diagnostic accuracy of the following three Models based on CAD-RADS classification: Model A (CAD-RADS 0 vs CAD-RADS 1–2 vs CAD-RADS 3,4,5), Model 1 (CAD-RADS 0 vs CAD-RADS>0), Model 2 (CAD-RADS 0–2 vs CAD-RADS 3–5). Time of analysis for both physicians and CNN were recorded. Model A showed a sensitivity, specificity, negative predictive value, positive predictive value and accuracy of 47%, 74%, 77%, 46% and 60%, respectively. Model 1 showed a sensitivity, specificity, negative predictive value, positive predictive value and accuracy of 66%, 91%, 92%, 63%, 86%, respectively. Conversely, Model 2 demonstrated the following sensitivity, specificity, negative predictive value, positive predictive value and accuracy: 82%, 58%, 74%, 69%, 71%, respectively. Time of analysis was significantly lower using CNN as compared to on-site reading (530.5 ± 179.1 vs 104.3 ± 1.4 sec, p =0.01) Deep CNN yielded accurate automated classification of patients with CAD-RADS. Image 1 • Deep convolutional neural network (CNN) yielded accurate automated Coronary Artery Disease Reporting and Data System (CAD-RADS) classification in patients with suspicious CAD. • CAD-RADS classification is significantly faster compared to human evaluation. • CNN can reduce the time of CCTA reporting in the next future. [ABSTRACT FROM AUTHOR]

Artificial intelligence (AI) in medical imaging is a potentially disruptive technology. An understanding of the principles and application of radiomics, artificial neural networks, machine learning, and deep learning is an essential foundation to weave design solutions that accommodate ethical and regulatory requirements, and to craft AI-based algorithms that enhance outcomes, quality, and efficiency. Moreover, a more holistic perspective of applications, opportunities, and challenges from a programmatic perspective contributes to ethical and sustainable implementation of AI solutions. [ABSTRACT FROM AUTHOR]

Effective fault diagnosis of rotating machinery has always been an important issue in real industries. In the recent years, data-driven fault diagnosis methods such as neural networks have been receiving increasing attention due to their great merits of high diagnosis accuracy and easy implementation. However, it is mostly difficult to fully train a deep neural network since gradients in optimization may vanish or explode during back-propagation, which results in deterioration and noticeable variance in model performance. In fault diagnosis researches, larger data sequence of machinery vibration signal containing sufficient information is usually preferred and consequently, deep models with large capacity are generally adopted. In order to improve network training, a residual learning algorithm is proposed in this paper. The proposed architecture significantly improves the information flow throughout the network, which is well suited for processing machinery vibration signal with variable sequential length. Little prior expertise on fault diagnosis and signal processing is required, that facilitates industrial applications of the proposed method. Experiments on a popular rolling bearing dataset are implemented to validate the proposed method. The results of this study suggest that the proposed intelligent fault diagnosis method for rotating machinery offers a new and promising approach. • Large neural network can be efficiently trained in fault diagnosis. • Machinery signal with variable sequential length can be processed. • Accurate testing diagnosis results are obtained with small variance. • Little prior expertise on signal processing and fault diagnosis is required. [ABSTRACT FROM AUTHOR]

Digital soil mapping (DSM) techniques are widely employed to generate soil maps. Soil properties are typically predicted individually, while ignoring the interrelation between them. Models for predicting multiple properties exist, but they are computationally demanding and often fail to provide accurate description of the associated uncertainty. In this paper a convolutional neural network (CNN) model is described to predict several soil properties with quantified uncertainty. CNN has the advantage that it incorporates spatial contextual information of environmental covariates surrounding an observation. A single CNN model can be trained to predict multiple soil properties simultaneously. I further propose a two-step approach to estimate the uncertainty of the prediction for mapping using a neural network model. The methodology is tested mapping six soil properties on the French metropolitan territory using measurements from the LUCAS dataset and a large set of environmental covariates portraying the factors of soil formation. Results indicate that the multivariate CNN model produces accurate maps as shown by the coefficient of determination and concordance correlation coefficient, compared to a conventional machine learning technique. For this country extent mapping, the maps predicted by CNN have a detailed pattern with significant spatial variation. Evaluation of the uncertainty maps using the median of the standardized squared prediction error and accuracy plots suggests that the uncertainty was accurately quantified, albeit slightly underestimated. The tests conducted using different window size of input covariates to predict the soil properties indicate that CNN benefits from using local contextual information in a radius of 4.5 km. I conclude that CNN is an effective model to predict several soil properties and that the associated uncertainty can be accurately quantified with the proposed approach. • A convolutional neural network is used for soil mapping. • Soil texture, organic carbon, pH and nitrogen are predicted using a single model. • The model is trained on topsoil LUCAS data over France. • A two-step method to quantify the uncertainty of the prediction is tested. • The method for uncertainty quantification is applicable to any neural network model. [ABSTRACT FROM AUTHOR]

Abstract Surface descriptors, which represent the surface characteristics of an image numerically, are the fundamental elements in many vision applications. Although traditional surface descriptors that are handcrafted or learned using machine learning techniques have been applied in many different vision applications, some difficulty remains in handling large amounts of noise and variance in 3D data. To resolve this difficulty, recent studies have applied deep learning techniques for the development of surface descriptors. Unlike other techniques based on the complete 3D CAD model or pre-known mesh information of the object, we consider the constraint of the robotic applications in which the information mentioned above is difficult to preload. In this paper, we propose a new 3D surface descriptor that does not require any pre-loaded topological information of the objects or a mesh construction, which may occasionally fail with new or previously unknown objects. Further, we propose a voxel representation that is adaptive to the density of the points, resolving the problem of varying densities of the point cloud data. Finally, we adopt domain-adversarial learning that leads a network to learn the features discriminative for similarity measurements while remaining invariant to different point densities. We gathered approximately 5,000 point-cloud images of objects along with their position and orientation information. We then constructed approximately half a million pairs of point clouds indicating the identical and different parts of the objects, which are labeled as true and false, respectively. The dataset of constructed pairs was used for the learning of 3D surface descriptors using a Siamese convolutional neural network (SCNN) with a domain-adversarial characteristic. The results indicate that the proposed descriptor outperforms other descriptors. Highlights • New point cloud dataset including images of identical scenes at three resolutions. • A study over algorithms that convert point cloud into voxel representation. • Feature learning by using domain-adversarial Siamese convolutional neural network. • Intensive evaluation of surface descriptors is performed on public datasets. [ABSTRACT FROM AUTHOR]

Highlights • Motion blur kernel size is a significant input parameter of deblurring algorithms. • A regression CNN is proposed to estimate the kernel size of a blurry image. • The kernel size for training is labeled in a relative way. • The kernel size to be estimated can be arbitrary. • The estimated kernel size helps existing deblurring algorithms yielding satisfactory results. Abstract Deblurring is to restore a latent clear image as well as to estimate an underlying blur kernel from a single blurry image. Motion blur kernel size is a significant input parameter of existing deblurring algorithms. Setting the size manually, which is a tedious trial-and-error process, can hardly obtain a satisfactory result directly and effectively. In this paper, we formulate the kernel size estimation as a regression problem and construct a convolutional neural network (CNN) to resolve it. After training the CNN by a large number of simulated blurry images with labels, which are produced by a relative way, we can estimate the kernel sizes of a given blurry image without size limitation. Both quantitative and qualitative results of experiments show that our method can not only estimate the motion blur kernel size accurately but also help to yield better pictures with estimated kernel sizes as input parameters. [ABSTRACT FROM AUTHOR]

This article deals with the agriculture, as part of the primary sector of the economy, includes the transformation of the natural environment through a set of actions and human intervention that allows satisfying the production mainly of food and derived raw materials. Therefore, with the high demand for products required for agricultural activity, it is necessary to implement new technologies to guarantee the quality and performance of production techniques and reduce environmental impact. In this context, precision agriculture has emerged to improve, evaluate, estimate and understand, based on the information obtained, the needs of crops. Taking into account the evolution of the methodology for processing of satellite images and the obtaining of indirect data, a classification of the characteristics of crop yield can be made. Definitely, different techniques have been applied to the processing of satellite images, but recently a current approach, both in effectiveness and speed in obtaining excellent results is the use of deep learning of the convolution of neural networks. The deep convolution of neural networks is used both in the recognition as well as in the classification of the satellite images of the sugarcane plantation in the Troncal region of the Coast of Ecuador. The experiment showed affirmative results of approximately 95% probability of recognition of crop status. [ABSTRACT FROM AUTHOR]

Abstract The stochastic gradient descent algorithm (SGD) is the main optimization solution in deep learning. The performance of SGD depends critically on how learning rates are tuned over time. In this paper, we propose a novel energy index based optimization method (EIOM) to automatically adjust the learning rate in the backpropagation. Since a frequently occurring feature is more important than a rarely occurring feature, we update the features to different extents according to their frequencies. We first define an energy neuron model and then design an energy index to describe the frequency of a feature. The learning rate is taken as a hyperparameter function according to the energy index. To empirically evaluate the EIOM, we investigate different optimizers with three popular machine learning models: logistic regression, multilayer perceptron, and convolutional neural network. The experiments demonstrate the promising performance of the proposed EIOM compared with that of other optimization algorithms. [ABSTRACT FROM AUTHOR]

Abstract Detecting in prior bearing faults is an essential task of machine health monitoring because bearings are the vital components of rotary machines. The performance of traditional intelligent fault diagnosis methods depend on feature extraction of fault signals, which requires signal processing techniques, expert knowledge, and human labor. Recently, deep learning algorithms have been applied widely in machine health monitoring. With the capacity of automatically learning complex features of input data, deep learning architectures have great potential to overcome drawbacks of traditional intelligent fault diagnosis. This paper proposes a method for diagnosing bearing faults based on a deep structure of convolutional neural network. Using vibration signals directly as input data, the proposed method is an automatic fault diagnosis system which does not require any feature extraction techniques and achieves very high accuracy and robustness under noisy environments. [ABSTRACT FROM AUTHOR]

Abstract A novel deep learning based eddy current inversion algorithm is proposed and investigated in this paper. Eddy current testing (ECT) for defects detection problem is adopted to demonstrated the proposed algorithms. The proposed model based on a Convolutional Neural Network (CNN) is developed to improve the defect detection performance with uncertainty information. The novelty of our work consists in combining characteristics of ECT data with general deep learning model to improve performance of deep learning in ECT field including a region of interest (ROI) method based on robust principle component analysis, a CNN classification model with weighted loss function and measurement of uncertainties. Experimental dataset obtained from eddy current inspection of heat exchanger tubes is utilized to validate the detection performance improvement. As a result, both the classification accuracy and the percentage of defects correctly identified have been increased to almost 100%. [ABSTRACT FROM AUTHOR]

Abstract Identifying and counting fish individuals on photos and videos is a crucial task to cost-effectively monitor marine biodiversity, yet it remains difficult and time-consuming. In this paper, we present a method to assist the identification of fish species on underwater images, and we compare our model performances to human ability in terms of speed and accuracy. We first tested the performance of a convolutional neural network (CNN) trained with different photographic databases while accounting for different post-processing decision rules to identify 20 fish species. Finally, we compared the performance of species identification of our best CNN model with that of humans on a test database of 1197 fish images representing nine species. The best CNN was the one trained with 900,000 images including (i) whole fish bodies, (ii) partial fish bodies and (iii) the environment (e.g. reef bottom or water). The rate of correct identification was 94.9%, greater than the rate of correct identification by humans (89.3%). The CNN was also able to identify fish individuals partially hidden behind corals or behind other fish and was more effective than humans to identify fish on smallest or blurry images while humans were better to identify fish individuals in unusual positions (e.g. twisted body). On average, each identification by our best CNN using a common hardware took 0.06 s. Deep Learning methods can thus perform efficient fish identification on underwater images and offer promises to build-up new video-based protocols for monitoring fish biodiversity cheaply and effectively. Highlights • Comparison between human experts and Deep Learning based method • Assessing the importance of post treatment with DL based methods • Importance of database adjustements • Real-life use cases for applied DL methods for fish identification [ABSTRACT FROM AUTHOR]

Abstract In attempt to improve the performance of a recently proposed hybrid human-machine interface, the eye-brain-computer interface (EBCI), we applied a compact convolutional neural network, the EEGNet, to short electroencephalogram (EEG) segments obtained during spontaneous and intentional gaze fixations, without the feature extraction step prior to classification. A statistically significant improvement of classification performance was obtained compared to with the results of the classifier previously used in the EBCI paradigm, which was based on shrinkage linear discriminant analysis (sLDA). Computation speed allows for using the EEGNet in the EBCI in online mode. [ABSTRACT FROM AUTHOR]

Graphical abstract Abstract Fast prediction of permeability directly from images enabled by image recognition neural networks is a novel pore-scale modeling method that has a great potential. This article presents a framework that includes (1) generation of porous media samples, (2) computation of permeability via fluid dynamics simulations, (3) training of convolutional neural networks (CNN) with simulated data, and (4) validations against simulations. Comparison of machine learning results and the ground truths suggests excellent predictive performance across a wide range of porosities and pore geometries, especially for those with dilated pores. Owning to such heterogeneity, the permeability cannot be estimated using the conventional Kozeny–Carman approach. Computational time was reduced by several orders of magnitude compared to fluid dynamic simulations. We found that, by including physical parameters that are known to affect permeability into the neural network, the physics-informed CNN generated better results than regular CNN. However, improvements vary with implemented heterogeneity. [ABSTRACT FROM AUTHOR]

Training a deep neural network with a large number of parameters often leads to overfitting problem. Recently, Dropout has been introduced as a simple, yet effective regularization approach to combat overfitting in such models. Although Dropout has shown remarkable results on many deep neural network cases, its actual effect on CNN has not been thoroughly explored. Moreover, training a Dropout model will significantly increase the training time as it takes longer time to converge than a non-Dropout model with the same architecture. To deal with these issues, we address Biased Dropout and Crossmap Dropout, two novel approaches of Dropout extension based on the behavior of hidden units in CNN model. Biased Dropout divides the hidden units in a certain layer into two groups based on their magnitude and applies different Dropout rate to each group appropriately. Hidden units with higher activation value, which give more contributions to the network final performance, will be retained by a lower Dropout rate, while units with lower activation value will be exposed to a higher Dropout rate to compensate the previous part. The second approach is Crossmap Dropout, which is an extension of the regular Dropout in convolution layer. Each feature map in a convolution layer has a strong correlation between each other, particularly in every identical pixel location in each feature map. Crossmap Dropout tries to maintain this important correlation yet at the same time break the correlation between each adjacent pixel with respect to all feature maps by applying the same Dropout mask to all feature maps, so that all pixels or units in equivalent positions in each feature map will be either dropped or active during training. Our experiment with various benchmark datasets shows that our approaches provide better generalization than the regular Dropout. Moreover, our Biased Dropout takes faster time to converge during training phase, suggesting that assigning noise appropriately in hidden units can lead to an effective regularization. [ABSTRACT FROM AUTHOR]

Spatio-temporal context (STC) based visual tracking algorithms have demonstrated remarkable tracking capabilities in recent years. In this paper, we propose an improved STC method, which seamlessly integrates capabilities of the powerful feature representations and mappings from the convolutional neural networks (CNNs) based on the theory of transfer learning. Firstly, the dynamic training confidence map, obtained from a mapping neural network using transferred CNN features, rather than the fixed training confidence map is utilized in our tracker to adapt the practical tracking scenes better. Secondly, we exploit hierarchical features from both the original image intensity and the transferred CNN features to construct context prior models. In order to enhance the accuracy and robustness of our tracker, we simultaneously transfer the fine-grained and semantic features from deep networks. Thirdly, we adopt the training confidence index (TCI), reflected from the dynamic training confidence map, to guide the updating process. It can determine whether back propagations should be conducted in the mapping neural network, and whether the STC model should be updated. The introduction of the dynamic training confidence map could effectively deal with the problem of location ambiguity further in our tracker. Overall, the comprehensive experimental results illustrate that the tracking capability of our tracker is competitive against several state-of-the-art trackers, especially the baseline STC tracker, on the existing OTB-2015 and UAV123 visual tracking benchmarks. [ABSTRACT FROM AUTHOR]

Abstract Convolutional neural networks (CNNs) have been a focus area of machine learning in recent years, and they are widely used in vision and speech processing because of their superior performance. However, CNNs are usually resource-heavy to ensure higher accuracy, i.e., an accurate network with millions of parameters requires high performance computing devices. This prevents the use of CNNs in resource-limited hardware. In this paper, we propose a novel CNN simplification method to prune feature maps with relatively low discriminability magnitudes, which can produce a simplified CNN with reduced computational cost. Specifically, we define the critical points among the discriminability values of feature maps in each convolutional layer, and use these critical points to easily find the best pruning number of feature maps. Our experimental results show that in each convolutional layer of the VGG model, 15.6% to 59.7% of feature maps can be pruned without any loss of accuracy in classification tasks. [ABSTRACT FROM AUTHOR]

Pedestrian detection and tracking have become an important field in the computer vision research area. This growing interest, started in the last decades, might be explained by the multitude of potential applications that could use the results of this research field, e.g. robotics, entertainment, surveillance, care for the elderly and disabled, and content-based indexing. In this survey paper, vision-based pedestrian detection systems are analysed based on their field of application, acquisition technology, computer vision techniques and classification strategies. Three main application fields have been individuated and discussed: video surveillance, human-machine interaction and analysis. Due to the large variety of acquisition technologies, this paper discusses both the differences between 2D and 3D vision systems, and indoor and outdoor systems. The authors reserved a dedicated section for the analysis of the Deep Learning methodologies, including the Convolutional Neural Networks in pedestrian detection and tracking, considering their recent exploding adoption for such a kind systems. Finally, focusing on the classification point of view, different Machine Learning techniques have been analysed, basing the discussion on the classification performances on different benchmark datasets. The reported results highlight the importance of testing pedestrian detection systems on different datasets to evaluate the robustness of the computed groups of features used as input to classifiers. [ABSTRACT FROM AUTHOR]

In recent years, advanced neurocomputing and machine learning techniques have been used for Electroencephalogram (EEG)-based diagnosis of various neurological disorders. In this paper, a novel computer model is presented for EEG-based screening of depression using a deep neural network machine learning approach, known as Convolutional Neural Network (CNN). The proposed technique does not require a semi-manually-selected set of features to be fed into a classifier for classification. It learns automatically and adaptively from the input EEG signals to differentiate EEGs obtained from depressive and normal subjects. The model was tested using EEGs obtained from 15 normal and 15 depressed patients. The algorithm attained accuracies of 93.5% and 96.0% using EEG signals from the left and right hemisphere, respectively. It was discovered in this research that the EEG signals from the right hemisphere are more distinctive in depression than those from the left hemisphere. This discovery is consistent with recent research and revelation that the depression is associated with a hyperactive right hemisphere. An exciting extension of this research would be diagnosis of different stages and severity of depression and development of a Depression Severity Index (DSI). [ABSTRACT FROM AUTHOR]

The most successful video-based human action recognition methods rely on feature representations extracted using Convolutional Neural Networks (CNNs). Inspired by the two-stream network (TS-Net), we propose a multi-stream Convolutional Neural Network (CNN) architecture to recognize human actions. We additionally consider human-related regions that contain the most informative features. First, by improving foreground detection, the region of interest corresponding to the appearance and the motion of an actor can be detected robustly under realistic circumstances. Based on the entire detected human body, we construct one appearance and one motion stream. In addition, we select a secondary region that contains the major moving part of an actor based on motion saliency. By combining the traditional streams with the novel human-related streams, we introduce a human-related multi-stream CNN (HR-MSCNN) architecture that encodes appearance, motion, and the captured tubes of the human-related regions. Comparative evaluation on the JHMDB, HMDB51, UCF Sports and UCF101 datasets demonstrates that the streams contain features that complement each other. The proposed multi-stream architecture achieves state-of-the-art results on these four datasets. [ABSTRACT FROM AUTHOR]

Recent years have witnessed the great success of convolutional neural network (CNN) based models in the field of computer vision. CNN is able to learn hierarchically abstracted features from images in an end-to-end training manner. However, most of the existing CNN models only learn features through a feedforward structure and no feedback information from top to bottom layers is exploited to enable the networks to refine themselves. In this paper, we propose a Learning with Rethinking algorithm. By adding a feedback layer and producing the emphasis vector, the model is able to recurrently boost the performance based on previous prediction. Particularly, it can be employed to boost any pre-trained models. This algorithm is tested on four object classification benchmark datasets: CIFAR-100, CIFAR-10, MNIST-background-image and ILSVRC-2012 dataset, and the results have demonstrated the advantage of training CNN models with the proposed feedback mechanism. [ABSTRACT FROM AUTHOR]

Access control is the act of providing privacy to a resource, and authentication through a single factor is no longer reliable to provide robust protection against unauthorized access. Hence, there is a rapid growth of exploring novel multi-factor authentication (MFA) methods which combine two or more authentication factors– inherence, possession, and knowledge. Despite the increasing use of MFA, to the best of authors’ knowledge, none have so far explored the combination of face, one-time password (OTP) and hand gesture in MFA. Thus, this study produces a proof-of-concept of this combination to form a new authentication method (Faceture ID). Furthermore, this study highlights three contributions: i) face verification with single-sample gallery set using pre-trained Deep Convolutional Neural Network, ii) handwriting gesture recognition using Leap Motion controller for tracking motion and Convolutional Neural Network for classification, and lastly, iii) a new MFA method utilizing face, OTP and hand gesture. The experimental results on the face verification show an average false acceptance rate of 1.94% with average genuine acceptance rate of 67.7%, from the self-built database where the facial images are exposed to variations in pose, expression, and occlusion. In addition, the classifier for the handwriting gesture recognition can predict gestures at about 96% for both precision and recall. Furthermore, the proposed MFA provides a novel systematic approach, high accuracy and performance with the intent to contribute on strengthening the security on privacy of resources against identity theft and attacks. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks have made great progress in recent handwritten character recognition (HCR) by learning discriminative features from large amounts of labeled data. However, the large variance of handwriting styles across writers is still a big challenge to the robust HCR. To alleviate this issue, an intuitional idea is to extract writer-independent semantic features from handwritten characters, while standard printed characters are writer-independent stencils for handwritten characters. They could be used as prior knowledge to guide models to exploit writer-independent semantic features for HCR. In this paper, we propose a novel adversarial feature learning (AFL) model to incorporate the prior knowledge of printed data and writer-independent semantic features to improve the performance of HCR on limited training data. Different from available handcrafted features methods, the proposed AFL model exploits writer-independent semantic features automatically, and standard printed data as prior knowledge is learnt objectively. Systematic experiments on MNIST and CASIA–HWDB show that the proposed model is competitive with the state-of-the-art methods on the offline HCR task. [ABSTRACT FROM AUTHOR]

Since Convolutional Neural Network (CNN) won the image classification competition 202 (ILSVRC12), a lot of attention has been paid to deep layer CNN study. The success of CNN is attributed to its superior multi-scale high-level image representations as opposed to hand-engineering low-level features. However, estimating millions of parameters of a deep CNN requires a large number of annotated samples, which currently prevents many superior deep CNNs (such as AlexNet, VGG, ResNet) being applied to problems with limited training data. To address this problem, a novel two-phase method combining CNN transfer learning and web data augmentation is proposed. With our method, the useful feature presentation of pre-trained network can be efficiently transferred to target task, and the original dataset can be augmented with the most valuable Internet images for classification. Our method not only greatly reduces the requirement of a large training data, but also effectively expand the training dataset. Both of method features contribute to the considerable over-fitting reduction of deep CNNs on small dataset. In addition, we successfully apply Bayesian optimization to solve the tuff problem, hyper-parameter tuning, in network fine-tuning. Our solution is applied to six public small datasets. Extensive experiments show that, comparing to traditional methods, our solution can assist the popular deep CNNs to achieve better performance. Particularly, ResNet can outperform all the state-of-the-art models on six small datasets. The experiment results prove that the proposed solution will be the great tool for dealing with practice problems which are related to use deep CNNs on small dataset. [ABSTRACT FROM AUTHOR]

Person re-identification (Re-ID) usually suffers from noisy samples with background clutter and mutual occlusion, which makes it extremely difficult to distinguish different individuals across the disjoint camera views. In this paper, we propose a novel deep self-paced learning (DSPL) algorithm to alleviate this problem, in which we apply a self-paced constraint and symmetric regularization to help the relative distance metric training the deep neural network, so as to learn the stable and discriminative features for person Re-ID. Firstly, we propose a soft polynomial regularizer term which can derive the adaptive weights to samples based on both the training loss and model age. As a result, the high-confidence fidelity samples will be emphasized and the low-confidence noisy samples will be suppressed at early stage of the whole training process. Such a learning regime is naturally implemented under a self-paced learning (SPL) framework, in which samples weights are adaptively updated based on both model age and sample loss using an alternative optimization method. Secondly, we introduce a symmetric regularizer term to revise the asymmetric gradient back-propagation derived by the relative distance metric, so as to simultaneously minimize the intra-class distance and maximize the inter-class distance in each triplet unit. Finally, we build a part-based deep neural network, in which the features of different body parts are first discriminately learned in the lower convolutional layers and then fused in the higher fully connected layers. Experiments on several benchmark datasets have demonstrated the superior performance of our method as compared with the state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Hyperspectral image classification is an important task in remote sensing image analysis. Traditional machine learning techniques are difficult to deal with hyperspectral images directly, because hyperspectral images have too many redundant spectral channels. In this paper we propose a novel method for hyperspectral image classification, by which spectral and spatial features are jointly exploited from hyperspectral images. Firstly, considering the local similarity in spatial domain, we employ a large spatial window to get image blocks from hyperspectral image Secondly, each spectral channel of the image block is filtered to extract their spatial and spectral features, after that the features are merged by convolutional layers. Finally, the fully-connected layers are used to get the classification result. Comparing with other state-of-the-art techniques, the proposed method pays more attention to the correlation of spatial neighborhood by using a large spatial window in the network. In addition, we combine the proposed network with the traditional support vector machine (SVM) classifier to improve the performance of hyperspectral image classification. Moreover, an adaptive method of the spatial window sizes selection is proposed in this paper. Experimental results conducted on the AVIRIS and ROSIS datasets demonstrate that the proposed method outperforms the state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

Automatic age and gender classification has been widely used in a large amount of applications, particularly in human-computer interaction, biometrics, visual surveillance, electronic customer, and commercial applications. In this paper, we introduce a hybrid structure which includes Convolutional Neural Network (CNN) and Extreme Learning Machine (ELM), and integrates the synergy of two classifiers to deal with age and gender classification. The hybrid architecture makes the most of their advantages: CNN is used to extract the features from the input images while ELM classifies the intermediate results. We not only give the detailed deployment of our structure including design of parameters and layers, analysis of the hybrid architecture, and the derivation of back-propagation in this system during the iterations, but also adopt several measures to limit the risk of overfitting. After that, two popular datasets, such as, MORPH-II and Adience Benchmark, are used to verify our hybrid structure. Experimental results show that our hybrid architecture outperforms other studies on the same datasets by exhibiting significant performance improvement in terms of accuracy and efficiency. [ABSTRACT FROM AUTHOR]

Different words in discourse arguments usually have varying contributions on the recognition of implicit discourse relations. Following this intuition, we propose two attention-based neural networks, namely inner attention model and outer attention model , to learn better discourse representation by automatically estimating the degrees of relevance of words to discourse relations. The former model only utilizes the information inside discourse arguments, while the latter model builds upon an outside semantic memory to exploit general world knowledge. Both models are capable of assigning more weights to relation-relevant words, and operate in an end-to-end manner. Upon these two models, we further propose a full attention model that combines their strengths into a unified framework. Extensive experiments on the PDTB data set show that our model significantly benefits from highlighting relation-relevant words and yields competitive and even better results against several state-of-the-art systems. [ABSTRACT FROM AUTHOR]

We propose a face alignment method that uses a deep neural network employing both local feature learning and recurrent regression. This method is primarily based on a convolutional neural network(CNN), which automatically learns local feature descriptors from the local facial landmark dataset that we created. Our research is motivated by the belief that investigating a face from its low-level component features would produce more competitive face alignment results, just as a CNN is normally trained to automatically learn a feature hierarchy from the lowest to the highest levels of abstraction. Moreover, by separately training the feature extraction layers and the regression layers, we impose an explicit functional discrimination between the feature extraction and regression tasks. First, we train a feature extraction network that is used to classify the landmark patches in the dataset. Using this pre-trained feature extraction network, we build a face alignment network, which uses an entire face image rather than the local landmark patch as input, thus generating the global facial features. The subsequent local feature extraction layer extracts the local feature set from this global feature, finally generating the local feature descriptors, in which space the network learns a generic descent direction from the currently estimated landmark positions to the ground truth via linear regression applied recurrently. Head pose estimation network also applied to provide a good initial estimate to the local feature extraction layer for accurate convergence. We found that learning of the good local landmark features in pursuit of good landmark classification also leads to a higher face alignment accuracy and achieves state-of-the-art performance on several public benchmark dataset. It signifies the importance of learning not only the global features but the local features for face alignment. We further verify our method's effectiveness when applied to related problems such as head pose estimation, facial landmark tracking, and invisible landmark detection. We believe that good local learning enables a deeper understanding of the face or object resulting in higher performance. [ABSTRACT FROM AUTHOR]

Recently, Convolutional Neural Networks (CNNs) have become a popular choice to tackle image classification tasks. Despite that, it is almost infeasible to embed the CNNs into resource limited hardware (e.g. mobile devices) due to its extremely high memory requirement. To address this problem, several methods were proposed to reduce the CNN memory requirement with a minimum compensation on the classification accuracy. In this paper, we propose a novel one-shot deep compression method based on the fuzzy quantity space to remove redundant CNN weights. Experiments in three public datasets (i.e. MNIST, CIFAR-10 and ImageNet) showed that our proposed approach is able to compress the CNN up to 14 × with a minimal loss of classification accuracy. Also, we present the first attempt to train an end-to-end fuzzy qualitative deep compression model in the fine-art paintings classification problem. We argue that the classification of fine-art collections is a more challenging problem in comparison to objects classification. This is because some of the artworks are neither non-representational nor figurative, and might even require imagination to recognize them. Hence, a question may arise as to whether a machine is able to capture imagination in paintings. One way to find out is by training a deep model and then visualize the low-level to high-level features learnt. Extensive experiments have been conducted on the recently publicly available Wikiart paintings dataset that consists of more than 80,000 paintings and our solution achieve state-of-the-art results (68%) in overall performance. The source code and models are available at: https://github.com/cs-chan/fuzzyDCN . [ABSTRACT FROM AUTHOR]

Emotion recognition from facial expression is the subfield of social signal processing which is applied in wide variety of areas, specifically for human and computer interaction. Many researches have been proposed for automatic emotion recognition, which is fundamentally using machine learning approach. However, recognizing basic emotions such as angry, happy, disgust, fear, sad, and surprise is still becoming a challenging problem in computer vision. Lately, deep learning has gained more attention to solve many real-world problems, including emotion recognition. In this research, we enhanced Convolutional Neural Network method to recognize 6 basic emotions and compared some preprocessing methods to show the influences of its in CNN performance. The compared data preprocessing methods are: resizing, face detection, cropping, adding noises, and data normalization consists of local normalization, global contrast normalization and histogram equalization. Face detection as single pre-processing phase achieved significant result with 86.08 % of accuracy, compared with another pre-processing phase and raw data. However, by combining those techniques can boost performance of CNN and achieved 97.06% of accuracy. [ABSTRACT FROM AUTHOR]

Recent studies on brain imaging analysis witnessed the core roles of machine learning techniques in computer-assisted intervention for brain disease diagnosis. Of various machine-learning techniques, sparse regression models have proved their effectiveness in handling high-dimensional data but with a small number of training samples, especially in medical problems. In the meantime, deep learning methods have been making great successes by outperforming the state-of-the-art performances in various applications. In this paper, we propose a novel framework that combines the two conceptually different methods of sparse regression and deep learning for Alzheimer’s disease/mild cognitive impairment diagnosis and prognosis. Specifically, we first train multiple sparse regression models, each of which is trained with different values of a regularization control parameter. Thus, our multiple sparse regression models potentially select different feature subsets from the original feature set; thereby they have different powers to predict the response values, i.e. , clinical label and clinical scores in our work. By regarding the response values from our sparse regression models as target-level representations , we then build a deep convolutional neural network for clinical decision making, which thus we call ‘ Deep Ensemble Sparse Regression Network .’ To our best knowledge, this is the first work that combines sparse regression models with deep neural network. In our experiments with the ADNI cohort, we validated the effectiveness of the proposed method by achieving the highest diagnostic accuracies in three classification tasks. We also rigorously analyzed our results and compared with the previous studies on the ADNI cohort in the literature. [ABSTRACT FROM AUTHOR]

Deep learning has received significant attention recently as a promising solution to many problems in the area of artificial intelligence. Among several deep learning architectures, convolutional neural networks (CNNs) demonstrate superior performance when compared to other machine learning methods in the applications of object detection and recognition. We use a CNN for image enhancement and the detection of driving lanes on motorways. In general, the process of lane detection consists of edge extraction and line detection. A CNN can be used to enhance the input images before lane detection by excluding noise and obstacles that are irrelevant to the edge detection result. However, training conventional CNNs requires considerable computation and a big dataset. Therefore, we suggest a new learning algorithm for CNNs using an extreme learning machine (ELM). The ELM is a fast learning method used to calculate network weights between output and hidden layers in a single iteration and thus, can dramatically reduce learning time while producing accurate results with minimal training data. A conventional ELM can be applied to networks with a single hidden layer; as such, we propose a stacked ELM architecture in the CNN framework. Further, we modify the backpropagation algorithm to find the targets of hidden layers and effectively learn network weights while maintaining performance. Experimental results confirm that the proposed method is effective in reducing learning time and improving performance. [ABSTRACT FROM AUTHOR]

Scene recognition plays an important role in the task of visual information retrieval, segmentation and image/video understanding. Traditional approaches for scene recognition usually utilize handcrafted features and have the drawbacks of poor representation ability, which can be improved by employing deep convolutional neural network (CNN) features that contain more semantic and structure information and thus possess more discriminative ability via multiple linear and non-linear transformations. However, an amount of detailed information may be lost when only the final output features which have gone through a certain number of transformations are applied to scene recognition. The features which are generated from the intermediate layers are not fully utilized. In this work, the GoogLeNet model is employed and divided into three parts of layers from bottom to top. The output features from each of the three parts are applied for scene recognition, which leads to the proposed GoogLeNet based multi-stage feature fusion (G-MS2F). What's more, the product rule is used to generate the final decision for scene recognition from the three outputs corresponding to the three parts of the proposed model. The experimental results demonstrate that the proposed model is superior to a number of state-of-the-art CNN models for scene recognition, and obtains the recognition accuracy of 92.90%, 79.63% and 64.06% on the benchmark scene recognition datasets Scene15, MIT67 and SUN397, respectively. [ABSTRACT FROM AUTHOR]

Computational color constancy refers to the problem of computing the illuminant color so that the images of a scene under varying illumination can be normalized to an image under the canonical illumination. In this paper, we adopt a deep learning framework for the illumination estimation problem. The proposed method works under the assumption of uniform illumination over the scene and aims for the accurate illuminant color computation. Specifically, we trained the convolutional neural network to solve the problem by casting the color constancy problem as an illumination classification problem. We designed the deep learning architecture so that the output of the network can be directly used for computing the color of the illumination. Experimental results show that our deep network is able to extract useful features for the illumination estimation and our method outperforms all previous color constancy methods on multiple test datasets. [ABSTRACT FROM AUTHOR]

High-resolution ultrasound is an emerging tool for diagnosing carpal tunnel syndrome caused by the compression of the median nerve at the wrist. This systematic review and meta-analysis aimed to explore and summarize the performance of deep learning algorithms in the automatic sonographic assessment of the median nerve at the carpal tunnel level. PubMed, Medline, Embase, and Web of Science were searched from the earliest records to May 2022 for studies investigating the utility of deep neural networks in the evaluation of the median nerve in carpal tunnel syndrome. The quality of the included studies was evaluated using the Quality Assessment Tool for Diagnostic Accuracy Studies. The outcome variables included precision, recall, accuracy, F-score, and Dice coefficient. In total, seven articles were included, comprising 373 participants. The deep learning and related algorithms comprised U-Net, phase-based probabilistic active contour, MaskTrack, ConvLSTM, DeepNerve, DeepSL, ResNet, Feature Pyramid Network, DeepLab, Mask R-CNN, region proposal network, and ROI Align. The pooled values of precision and recall were 0.917 (95 % confidence interval [CI], 0.873–0.961) and 0.940 (95 % CI, 0.892–0.988), respectively. The pooled accuracy and Dice coefficient were 0.924 (95 % CI, 0.840–1.008) and 0.898 (95 % CI, 0.872–0.923), respectively, whereas the summarized F-score was 0.904 (95 % CI, 0.871–0.937). The deep learning algorithm enables automated localization and segmentation of the median nerve at the carpal tunnel level in ultrasound imaging with acceptable accuracy and precision. Future research is expected to validate the performance of deep learning algorithms in detecting and segmenting the median nerve along its entire length as well as across datasets obtained from various ultrasound manufacturers. • This meta-analysis was the first to investigate the deep learning algorithm on the localization and segmentation of the median nerve near the carpal tunnel. • Our investigation revealed the potential of deep learning algorithm on motion metrics measurement of the median nerve during finger motion. • The impact of anatomic variance on detection of the median nerve was also investigated in our reviewed article. [ABSTRACT FROM AUTHOR]