Recognition of Film Type Using HSV Features on Deep-Learning Neural Networks

The number of films is numerous and the film contents are complex over the Internet and multimedia sources. It is time consuming for a viewer to select a favorite film. This paper presents an automatic recognition system of film types. Initially, a film is firstly sampled as frame sequences. The color space, including hue, saturation,and brightness value(HSV), is analyzed for each sampled frame by computing the deviation and mean of HSV for each film. These features are utilized as inputs to a deep-learning neural network(DNN) for the recognition of film types. One hundred films are utilized to train and validate the model parameters of DNN. In the testing phase, a film is recognized as one of the five categories, including action, comedy, horror thriller, romance, and science fiction, by the trained DNN. The experimental results reveal that the film types can be effectively recognized by the proposed approach, enabling the viewer to select an interesting film accurately and quickly.

A unified deep-learning network to accurately segment insulin granules of different animal models imaged under different electron microscopy methodologies

Insulin is important for body metabolism regulation and glucose homeostasis,and its dysregulation often leads to metabolic syndrome(MS)and diabetes.Insulin is normally stored in large dense-core vesicles(LDCVs)in pancreatic beta cells,and significant reductions in the number,size,gray level and density of insulin granules confer diabetes both in mice(Xue et al.,2012)and humans(Masini et al.,2012).

Hybrid pedestrian positioning system using wearable inertial sensors and ultrasonic ranging

Pedestrian positioning system(PPS)using wearable inertial sensors has wide applications towards various emerging fields such as smart healthcare,emergency rescue,soldier positioning,etc.The performance of traditional PPS is limited by the cumulative error of inertial sensors,complex motion modes of pedestrians,and the low robustness of the multi-sensor collaboration structure.This paper presents a hybrid pedestrian positioning system using the combination of wearable inertial sensors and ultrasonic ranging(H-PPS).A robust two nodes integration structure is developed to adaptively combine the motion data acquired from the single waist-mounted and foot-mounted node,and enhanced by a novel ellipsoid constraint model.In addition,a deep-learning-based walking speed estimator is proposed by considering all the motion features provided by different nodes,which effectively reduces the cumulative error originating from inertial sensors.Finally,a comprehensive data and model dual-driven model is presented to effectively combine the motion data provided by different sensor nodes and walking speed estimator,and multi-level constraints are extracted to further improve the performance of the overall system.Experimental results indicate that the proposed H-PPS significantly improves the performance of the single PPS and outperforms existing algorithms in accuracy index under complex indoor scenarios.

Robust Vehicle Detection Based on Improved You Look Only Once

Vehicle detection is still challenging for intelligent transportation systems(ITS)to achieve satisfactory performance.The existing methods based on one stage and two-stage have intrinsic weakness in obtaining high vehicle detection performance.Due to advancements in detection technology,deep learning-based methods for vehicle detection have become more popular because of their higher detection accuracy and speed than the existing algorithms.This paper presents a robust vehicle detection technique based on Improved You Look Only Once(RVD-YOLOv5)to enhance vehicle detection accuracy.The proposed method works in three phases;in the first phase,the K-means algorithm performs data clustering on datasets to generate the classes of the objects.Subsequently,in the second phase,the YOLOv5 is applied to create the bounding box,and the Non-Maximum Suppression(NMS)technique is used to eliminate the overlapping of the bounding boxes of the vehicle.Then,the loss function CIoU is employed to obtain the accurate regression bounding box of the vehicle in the third phase.The simulation results show that the proposed method achieves better results when compared with other state-of-art techniques,namely LightweightDilated Convolutional Neural Network(LD-CNN),Single Shot Detector(SSD),YOLOv3 and YOLOv4 on the performance metric like precision,recall,mAP and F1-Score.The simulation and analysis are carried out on PASCAL VOC 2007,2012 and MS COCO 2017 datasets to obtain better performance for vehicle detection.Finally,the RVD-YOLOv5 obtains the results with an mAP of 98.6%and Precision,Recall,and F1-Score are 98%,96.2%and 97.09%,respectively.

Intrusion Detection System with Customized Machine Learning Techniques for NSL-KDD Dataset

Modern networks are at risk from a variety of threats as a result of the enormous growth in internet-based traffic.By consuming time and resources,intrusive traffic hampers the efficient operation of network infrastructure.An effective strategy for preventing,detecting,and mitigating intrusion incidents will increase productivity.A crucial element of secure network traffic is Intrusion Detection System(IDS).An IDS system may be host-based or network-based to monitor intrusive network activity.Finding unusual internet traffic has become a severe security risk for intelligent devices.These systems are negatively impacted by several attacks,which are slowing computation.In addition,networked communication anomalies and breaches must be detected using Machine Learning(ML).This paper uses the NSL-KDD data set to propose a novel IDS based on Artificial Neural Networks(ANNs).As a result,the ML model generalizes sufficiently to perform well on untried data.The NSL-KDD dataset shall be utilized for both training and testing.In this paper,we present a custom ANN model architecture using the Keras open-source software package.The specific arrangement of nodes and layers,along with the activation functions,enhances the model’s ability to capture intricate patterns in network data.The performance of the ANN is carefully tested and evaluated,resulting in the identification of a maximum detection accuracy of 97.5%.We thoroughly compared our suggested model to industry-recognized benchmark methods,such as decision classifier combinations and ML classifiers like k-Nearest Neighbors(KNN),Deep Learning(DL),Support Vector Machine(SVM),Long Short-Term Memory(LSTM),Deep Neural Network(DNN),and ANN.It is encouraging to see that our model consistently outperformed each of these tried-and-true techniques in all evaluations.This result underlines the effectiveness of the suggested methodology by demonstrating the ANN’s capacity to accurately assess the effectiveness of the developed strategy in identifying and categorizing instances of network intrusion.

Deep learning method for cell count from transmitted-light microscope

Automatic cell counting provides an effective tool for medical research and diagnosis.Currently,cell counting can be completed by transmitted-light microscope,however,it requires expert knowledge and the counting accuracy which is unsatisfied for overlapped cells.Further,the image-translation-based detection method has been proposed and the potential has been shown to accomplish cell counting from transmitted-light microscope,automatically and effectively.In this work,a new deep-learning(DL)-based two-stage detection method(cGAN-YOLO)is designed to further enhance the performance of cell counting,which is achieved by combining a DL-based fluorescent image translation model and a DL-based cell detection model.The various results show that cGAN-YOLO can effectively detect and count some different types of cells from the acquired transmitted-light microscope images.Compared with the previously reported YOLO-based one-stage detection method,high recognition accuracy(RA)is achieved by the cGAN-YOLO method,with an improvement of 29.80%.Furthermore,we can also observe that cGAN-YOLO obtains an improvement of 12.11%in RA compared with the previously reported image-translation-based detection method.In a word,cGAN-YOLO makes it possible to implement cell counting directly from the experimental acquired transmitted-light microscopy images with high flexibility and performance,which extends the applicability in clinical research.

Application of artificial intelligence in gastroenterology

Artificial intelligence(AI) using deep-learning(DL) has emerged as a breakthrough computer technology. By the era of big data, the accumulation of an enormous number of digital images and medical records drove the need for the utilization of AI to efficiently deal with these data, which have become fundamental resources for a machine to learn by itself. Among several DL models, the convolutional neural network showed outstanding performance in image analysis. In the field of gastroenterology, physicians handle large amounts of clinical data and various kinds of image devices such as endoscopy and ultrasound. AI has been applied in gastroenterology in terms of diagnosis,prognosis, and image analysis. However, potential inherent selection bias cannot be excluded in the form of retrospective study. Because overfitting and spectrum bias(class imbalance) have the possibility of overestimating the accuracy,external validation using unused datasets for model development, collected in a way that minimizes the spectrum bias, is mandatory. For robust verification,prospective studies with adequate inclusion/exclusion criteria, which represent the target populations, are needed. DL has its own lack of interpretability.Because interpretability is important in that it can provide safety measures, help to detect bias, and create social acceptance, further investigations should be performed.

Predicting oral disintegrating tablet formulations by neural network techniques

Oral disintegrating tablets(ODTs) are a novel dosage form that can be dissolved on thetongue within 3 min or less especially for geriatric and pediatric patients. Current ODT for-mulation studies usually rely on the personal experience of pharmaceutical experts andtrial-and-error in the laboratory, which is inefficient and time-consuming. The aim of cur-rent research was to establish the prediction model of ODT formulations with direct com-pression process by artificial neural network(ANN) and deep neural network(DNN) tech-niques. 145 formulation data were extracted from Web of Science. All datasets were dividedinto three parts: training set(105 data), validation set(20) and testing set(20). ANN andDNN were compared for the prediction of the disintegrating time. The accuracy of the ANNmodel have reached 85.60%, 80.00% and 75.00% on the training set, validation set and testingset respectively, whereas that of the DNN model were 85.60%, 85.00% and 80.00%, respec-tively. Compared with the ANN, DNN showed the better prediction for ODT formulations.It is the first time that deep neural network with the improved dataset selection algorithmis applied to formulation prediction on small data. The proposed predictive approach couldevaluate the critical parameters about quality control of formulation, and guide researchand process development. The implementation of this prediction model could effectivelyreduce drug product development timeline and material usage, and proactively facilitatethe development of a robust drug product.

Automated classification of dual channel dental imaging of auto-fluorescence and white lightby convolutional neural networks

Prevention is the most effective way to reduce dental caries.In order to provide a simple way to achieve oral healthcare direction in daily life,dual Channel,portable dental Imaging system that combine white light with autofluorescence techniques was established,and then,a group of volunteers were recruited,7200 tooth pictures of different dental caries stage and dental plaque were taken and collected.In this work,a customized Convolutional Neural Networks(CNNs)have been designed to classify dental image with early stage caries and dental plaque.Eighty percentage(n=6000)of the pictures taken were used to supervised training of the CNNs based on the experienced dentists'advice and the rest 20%(n=1200)were used to a test dataset to test the trained CNNs.The accuracy,sensitivity and specificity were calculated to evaluate perfor-mance of the CNNs.The accuracy for the early stage caries and dental plaque were 95.3%and 95.9%,respectively.These results shown that the designed image system combined the cus-tomized CNNs that could automatically and efficiently find early caries and dental plaque on occlusal,lingual and buccal surfaces.Therefore,this will provide a novel approach to dental caries prevention for everyone in daily life.

An Innovative Bias-Correction Approach to CMA-GD Hourly Quantitative Precipitation Forecasts

This paper proposes a simple and powerful optimal integration(OPI)method for improving hourly quantitative precipitation forecasts(QPFs,0-24 h)of a single-model by integrating the benefits of different biascorrected methods using the high-resolution CMA-GD model from the Guangzhou Institute of Tropical and Marine Meteorology of China Meteorological Administration(CMA).Three techniques are used to generate multi-method calibrated members for OPI:deep neural network(DNN),frequency-matching(FM),and optimal threat score(OTS).The results are as follows:(1)The QPF using DNN follows the basic physical patterns of CMA-GD.Despite providing superior improvements for clear-rainy and weak precipitation,DNN cannot improve the predictions for severe precipitation,while OTS can significantly strengthen these predictions.As a result,DNN and OTS are the optimal members to be incorporated into OPI.(2)Our new approach achieves state-of-the-art performances on a single model for all magnitudes of precipitation.Compared with the CMA-GD,OPI improves the TS by 2.5%,5.4%,7.8%,8.3%,and 6.1%for QPFs from clear-rainy to rainstorms in the verification dataset.Moreover,OPI shows good stability in the test dataset.(3)It is also noted that the rainstorm pattern of OPI relies heavily on the original model and that OPI cannot correct for deviations in the location of severe precipitation.Therefore,improvements in predicting severe precipitation using this method should be further realized by improving the numerical model's forecasting capability.

Multi-Person Device-Free Gesture Recognition Using mmWave Signals

Device-free gesture recognition is an emerging wireless sensing technique which could recognize gestures by analyzing its influence on surrounding wireless signals,it may empower wireless networks with the augmented sensing ability.Researchers have made great achievements for singleperson device-free gesture recognition.However,when multiple persons conduct gestures simultaneously,the received signals will be mixed together,and thus traditional methods would not work well anymore.Moreover,the anonymity of persons and the change in the surrounding environment would cause feature shift and mismatch,and thus the recognition accuracy would degrade remarkably.To address these problems,we explore and exploit the diversity of spatial information and propose a multidimensional analysis method to separate the gesture feature of each person using a focusing sensing strategy.Meanwhile,we also present a deep-learning based robust device free gesture recognition framework,which leverages an adversarial approach to extract robust gesture feature that is insensitive to the change of persons and environment.Furthermore,we also develop a 77GHz mmWave prototype system and evaluate the proposed methods extensively.Experimental results reveal that the proposed system can achieve average accuracies of 93%and 84%when 10 gestures are conducted in Received:Jun.18,2020 Revised:Aug.06,2020 Editor:Ning Ge different environments by two and four persons simultaneously,respectively.

Super-Resolution Imaging of Mammograms Based on the Super-Resolution Convolutional Neural Network

Purpose: To apply and evaluate a super-resolution scheme based on the super-resolution convolutional neural network (SRCNN) for enhancing image resolution in digital mammograms. Materials and Methods: A total of 711 mediolateral oblique (MLO) images including breast lesions were sampled from the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM). We first trained the super-resolution convolutional neural network (SRCNN), which is a deep-learning based super-resolution method. Using this trained SRCNN, high-resolution images were reconstructed from low-resolution images. We compared the image quality of the super-resolution method and that obtained using the linear interpolation methods (nearest neighbor and bilinear interpolations). To investigate the relationship between the image quality of the SRCNN-processed images and the clinical features of the mammographic lesions, we compared the image quality yielded by implementing the SRCNN, in terms of the breast density, the Breast Imaging-Reporting and Data System (BI-RADS) assessment, and the verified pathology information. For quantitative evaluation, peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) were measured to assess the image restoration quality and the perceived image quality. Results: The super-resolution image quality yielded by the SRCNN was significantly higher than that obtained using linear interpolation methods (p p Conclusion: SRCNN can significantly outperform conventional interpolation methods for enhancing image resolution in digital mammography. SRCNN can significantly improve the image quality of magnified mammograms, especially in dense breasts, high-risk BI-RADS assessment groups, and pathology-verified malignant cases.

A novel physics-informed framework for reconstruction of structural defects

The ultrasonic guided wave technology plays a significant role in the field of non-destructive testing as it employs acoustic waves with the advantages of high propagation efficiency and low energy consumption during the inspect process.However,the theoretical solutions to guided wave scattering problems with assumptions such as the Born approximation have led to the poor quality of the reconstructed results.Besides,the scattering signals collected from industry sectors are often noised and nonstationary.To address these issues,a novel physics-informed framework(PIF)for the quantitative reconstruction of defects by means of the integration of the data-driven method with the guided wave scattering analysis is proposed in this paper.Based on the geometrical information of defects and initial results obtained by the PIF-based analysis of defect reconstructions,a deep-learning neural network model is built to reveal the physical relationship between the defects and the noisy detection signals.This learning model is then adopted to assess and characterize the defect profiles in structures,improve the accuracy of the analytical model,and eliminate the impact of the noise pollution in the process of inspection.To demonstrate the advantages of the developed PIF for the complex defect reconstructions with the capability of denoising,several numerical examples are carried out.The results show that the PIF has greater accuracy for the reconstruction of defects in the structures than the analytical method,and provides a valuable insight into the development of artificial intelligence(AI)-assisted inspection systems with high accuracy and efficiency in the fields of structural integrity and condition monitoring.

Deep Learning-based Wireless Signal Classification in the IoT Environment

With the development of the Internet of Things(IoT),diverse wireless devices are increasing rapidly.Those devices have different wireless interfaces that generate incompatible wireless signals.Each signal has its own physical characteristics with signal modulation and demodulation scheme.When there exist different wireless devices,they can suffer from severe Cross-Technology Interferences(CTI).To reduce the communication overhead due to the CTI in the real IoT environment,a central coordinator can be able to detect and identify wireless signals existing in the same communication areas.This paper investigates how to classify various radio signals using Convolutional Neural Networks(CNN),Long Short-TermMemory(LSTM)and attention mechanism.CNN can reduce the amount of computation by reducing weights by using convolution,and LSTM belonging to RNNmodels can alleviate the long-term dependence problem.Furthermore,attention mechanism can reduce the short-term memory problem of RNNs by reexamining the data output from the decoder and the entire data entered into the encoder at every point in time.To accurately classify radio signals according to their weights,we design a model based on CNN,LSTM,and attention mechanism.As a result,we propose a model CLARINet that can classify original data by minimizing the loss and detects changes in sequences.In a case of the real IoT environment with Wi-Fi,Bluetooth and ZigBee devices,we can normally obtain wireless signals from 10 to 20 dB.The accuracy of CLARINet’s radio signal classification with CNN-LSTM and attention mechanism can be seen that signal-to-noise ratio(SNR)exhibits high accuracy at 16 dB to about 92.03%.

Real-Time Anomaly Detection in Packaged Food X-Ray Images Using Supervised Learning

Physical contamination of food occurs when it comes into contact with foreign objects.Foreign objects can be introduced to food at any time during food delivery and packaging and can cause serious concerns such as broken teeth or choking.Therefore,a preventive method that can detect and remove foreign objects in advance is required.Several studies have attempted to detect defective products using deep learning networks.Because it is difficult to obtain foreign object-containing food data from industry,most studies on industrial anomaly detection have used unsupervised learning methods.This paper proposes a new method for real-time anomaly detection in packaged food products using a supervised learning network.In this study,a realistic X-ray image training dataset was constructed by augmenting foreign objects with normal product images in a cut-paste manner.Based on the augmented training dataset,we trained YOLOv4,a real-time object detection network,and detected foreign objects in the test data.We evaluated this method on images of pasta,snacks,pistachios,and red beans under the same conditions.The results show that the normal and defective products were classified with an accuracy of at least 94%for all packaged foods.For detecting foreign objects that are typically difficult to detect using the unsupervised learning and traditional methods,the proposed method achieved high-performance realtime anomaly detection.In addition,to eliminate the loss in high-resolution X-ray images,the false positive rate and accuracy could be lowered to 5%with patch-based training and a new post-processing algorithm.

Shadow Detection and Removal From Photo-Realistic Synthetic Urban Image Using Deep Learning

Recently,virtual reality technology that can interact with various data is used for urban design and analysis.Reality,one of the most important elements in virtual reality technology,means visual expression so that a person can experience three-dimensional space like reality.To obtain this realism,real-world data are used in the various fields.For example,in order to increase the realism of 3D modeled building textures real aerial images are utilized in 3D modelling.However,the aerial image captured during the day can be shadowed by the sun and it can cause the distortion or deterioration of image.To resolve this problem,researches on detecting and removing shadows have been conducted,but the detecting and removing shadow is still considered as a challenging problem.In this paper,we propose a novel method for detecting and removing shadows using deep learning.For this work,we first a build a new dataset of photo-realistic synthetic urban data based on the virtual environment using 3D spatial information provided by VWORLD.For detecting and removing shadow from the dataset,firstly,the 1-channel shadow mask image is inferred from the 3-channel shadow image through the CNN.Then,to generate a shadow-free image,a 3-channel shadow image and a detected 1-channel shadow mask into the GAN is executed.From the experiments,we can prove that the proposed method outperforms the existing methods in detecting and removing shadow.

Auto-Segmentation on Liver with U-Net and Pixel De-Convolutional Network

Purpose: To improve the liver auto-segmentation performance of three-dimensional (3D) U-net by replacing the conventional up-sampling convolution layers with the Pixel De-convolutional Network (PDN) that considers spatial features. Methods: The U-net was originally developed to segment neuronal structure with outstanding performance but suffered serious artifacts from indirectly unrelated adjacent pixels in its up-sampling layers. The hypothesis of this study was that the segmentation quality of the liver could be improved with PDN in which the up-sampling layer was replaced by a pixel de-convolution layer (PDL). Seventy-eight plans of abdominal cancer patients were anonymized and exported. Sixty-two were chosen for training two networks: 1) 3D U-Net, and 2) 3D PDN, by minimizing the Dice loss function. The other sixteen plans were used to test the performance. The similarity Dice and Average Hausdorff Distance (AHD) were calculated and compared between these two networks. Results: The computation time for 62 training cases and 200 training epochs was about 30 minutes for both networks. The segmentation performance was evaluated using the remaining 16 cases. For the Dice score, the mean ± standard deviation were 0.857 ± 0.011 and 0.858 ± 0.015 for the PDN and U-Net, respectively. For the AHD, the mean ± standard deviation were 1.575 ± 0.373 and 1.675 ± 0.769, respectively, corresponding to an improvement of 6.0% and 51.5% of mean and standard deviation for the PDN. Conclusion: The PDN has outperformed the U-Net on liver auto-segmentation. The predicted contours of PDN are more conformal and smoother when compared with the U-Net.

It from Qutrit: Braided Loop Metaheuristic

Cosmologies are developed by physicists and philosophers to explain our experiences of the evolving cosmos. Intelligent deep-learning metaheuristics provide original frameworks for cosmologies which are founded on quantum information. Mathematical standard models of physical cosmology and particle physics formalize an abundance of observations, yet there is no scientific consensus about how these models include our conscious experiences and fundamental philosophies of information. Furthermore, Naturalness in physics is coupled to the related problem of fine-tuning. To address these foundational problems, within the quantum information paradigm, whilst aligning with standard scientific models, I introduce a topological deep-learning cosmology metaheuristic. Braided, 3-coloured, world-strands are proposed to be the fundamental quantum information tracts (ethereal fibre bundles) of our evolving Triuniverse. This Braided Loop Metaheuristic comprises eternally evolving deep-learning feedback loops of superposed, braided, 3-coloured, quantum information world-strands, which process (in 3-level qutrit states) foundational properties coined Algebrus (labelled red), Algorithmus (labelled green) and Geometrus (labelled blue). Braids split from 1→2→3 (in knot representation respectively: closed loop→trefoil knot→Borromean loops) thence combine from 3→2→1 to form eternally evolving deep-learning loops. This cosmology metaheuristic simultaneously incorporates initial Laws of Form;Emergentism (from substrate Mathematics, through Quantum Physics to Life);Consciousness (as a superposed triunity of Implicate Order, Process Philosophy and Aesthetic Relationalism);Reductionism (from Life, through Quantum Physics to Pure Mathematics expressed as Logical Axioms, Laws of Parsimony and Ideal Form);and the Braided Loop Metaheuristic reboots its eternal cycle with the initial Laws of Form. An agent’s personal anthropic Braided Loop Metaheuristic represents one of many-worlds, a meridional loop in a multiverse with horn-torus topology, where Nature’s physical parameters vary equatorially. Fundamental information processing is driven by ψ-Epistemic Drive, the Natural appetite for information selected for advantageous knowledge. The meridional loops are ψ-Epistemic Field lines emanating from an epistemic dipole at the horn-torus core. Equatorial parameter fine-tuning in many-worlds quantum physics and the many-species of Darwinian Life are similar deep-learning optimizations in the Braided Loop Metaheuristic.

Artificial intelligence:Advances and new frontiers in medical imaging

Artificial intelligence(AI)has been entwined with the field of radiology ever since digital imaging began replacing films over half a century ago.These algorithms,ranging from simplistic speech-to-text dictation programs to automated interpretation neural networks,have continuously sought to revolutionize medical imaging.With the number of imaging studies outpacing the amount of trained of readers,AI has been implemented to streamline workflow efficiency and provide quantitative,standardized interpretation.AI relies on massive amounts of data for its algorithms to function,and with the wide-spread adoption of Picture Archiving and Communication Systems(PACS),imaging data is accumulating rapidly.Current AI algorithms using machine-learning technology,or computer aided-detection,have been able to successfully pool this data for clinical use,although the scope of these algorithms remains narrow.Many systems have been developed to assist the workflow of the radiologist through PACS optimization and imaging study triage,however interpretation has generally remained a human responsibility for now.In this review article,we will summarize the current successes and limitations of AI in radiology,and explore the exciting prospects that deep-learning technology offers for the future.

Interpreting random fields through the U-Net architecture for failure mechanism and deformation predictions of geosystems

The representation of spatial variation of soil properties in the form of random fields permits advanced probabilistic assessment of slope stability.In many studies,the safety margin of the system is typically characterized by the term“probability of failure(Pfailure)”.As the intensity and spatial distribution of soil properties vary in different random field realizations,the failure mechanism and deformation field of a slope can vary as well.Not only can the location of the failure surfaces vary,but the mode of failure also changes.Such information is equally valuable to engineering practitioners.In this paper,two slope examples that are modified from a real case study are presented.The first example pertains to the stability analysis of a multi-layer-slope while the second example deals with the serviceability analysis of a multi-layer c-φslope.In addition,due to the large number of simulations needed to reveal the full picture of the failure mechanism,Convolutional Neural Networks(CNNs)that adopt a U-Net architecture is proposed to offer a soft computing strategy to facilitate the investigation.The spatial distribution of the failure surfaces,the statistics of the sliding volume,and the statistics of the deformation field are presented.The results also show that the proposed deep-learning model is effective in predicting the failure mechanism and deformation field of slopes in spatially variable soils;therefore encouraging probabilistic study of slopes in practical scenarios.