Extract Salient Features Research Articles

A Fusion-Based Hybrid-Feature Approach for Recognition of Unconstrained Offline Handwritten Hindi Characters

Hindi is the official language of India and used by a large population for several public services like postal, bank, judiciary, and public surveys. Efficient management of these services needs language-based automation. The proposed model addresses the problem of handwritten Hindi character recognition using a machine learning approach. The pre-trained DCNN models namely; InceptionV3-Net, VGG19-Net, and ResNet50 were used for the extraction of salient features from the characters’ images. A novel approach of fusion is adopted in the proposed work; the DCNN-based features are fused with the handcrafted features received from Bi-orthogonal discrete wavelet transform. The feature size was reduced by the Principal Component Analysis method. The hybrid features were examined with popular classifiers namely; Multi-Layer Perceptron (MLP) and Support Vector Machine (SVM). The recognition cost was reduced by 84.37%. The model achieved significant scores of precision, recall, and F1-measure—98.78%, 98.67%, and 98.69%—with overall recognition accuracy of 98.73%.

Feature refinement: An expression-specific feature learning and fusion method for micro-expression recognition

Micro-expression recognition has become challenging, as it is extremely difficult to extract the subtle facial changes of micro-expressions. Recently, several approaches have proposed various expression-shared features algorithms for micro-expression recognition. However, these approaches do not reveal the specific discriminative characteristics, which leads to sub-optimal performance. This paper proposes a novel Feature Refinement (FeatRef) with expression-specific feature learning and fusion for micro-expression recognition that aims to obtain salient and discriminative features for specific expressions and predicts expressions by fusing expression-specific features. FeatRef consists of an expression proposal module with an attention mechanism and a classification branch. First, an inception module is designed based on optical flow to obtain expression-shared features. Second, to extract salient and discriminative features for specific expressions, expression-shared features are fed into an expression proposal module with attention factors and proposal loss. Last, in the classification branch, category labels are predicted via a fusion of expression-specific features. Experiments on three publicly available databases validate the effectiveness of FeatRef under different protocols. The results on public benchmarks demonstrate that FeatRef provides salient and discriminative information for micro-expression recognition. The results also show that FeatRef achieves better or competitive performance with existing state-of-the-art methods on micro-expression recognition.

Hyperspectral image classification based on spatial and spectral kernels generation network

With the widespread use of deep learning methods, more and more classification models based on hyperspectral images (HSI) have been continuously proposed. However, due to the characteristics of high dimensionality and low resolution of HSI, the traditional convolutional neural network (CNN) model cannot effectively process it. In this paper, we propose a classification network, called spatial and spectral kernels generation network (SSKNet), to generate spatial and spectral convolution kernels based on image characteristics, and use them to replace the traditional initialization convolution kernels. We can obtain representative kernels with stronger spatial correlation with the region segmentation, clustering, and mapping operations of the spatial kernels generation module (SaKG Module). Simultaneously, we also propose a spectral kernels generation module (SeKG module), which integrates the multi-scale correlation characteristics of different bands into the spectral attention mechanism, making the generated spectral kernels more accurate. Combining the spatial and spectral kernels allows the network to extract salient features and achieve feature fusion efficiently. Experimental results based on multiple HSI data sets show that the proposed method has better classification accuracy and generalization performance than existing methods.

Fusion of medical images based on salient features extraction by PSO optimized fuzzy logic in NSST domain

The aim of multimodal medical image fusion is to extract information from different modal images into a single one so that the single fusion image contains the salient features of the source images to the maximum extent. Different medical imaging modalities such as CT and MRI are presented as different visual morphology, which often show different complementary salient features in clinical diagnosis. According to this characteristic, in order to fuse different salient features of multimodal medical images into a single image and provide a new dimension of information for clinical diagnosis to improve the diagnosis accuracy, an improved image fusion algorithm based on visual salience detection is proposed in this paper. The visual salience of two registered source images is calculated by the GBVS algorithm, and the low-frequency and high-frequency sub-bands are obtained by decomposing the source images in NSST domain. For the low-frequency sub-bands, fuzzy logic system is used to obtain the respective weights of the fused low-frequency sub-band with the local energy and GBVS graph as input. For the high-frequency sub-bands, the NSML values of each sub-band are calculated and compared to obtain the fused high-frequency sub-band. The final fused image is obtained by the inverse NSST transformation. In addition, PSO algorithm is used to optimize the membership function of fuzzy logic system for better adaption to medical images and feature extraction. By applying this multimodal medical image fusion method, the visual quality of the image is effectively improved and the salient features of tissues are preserved well. Experiments on fusion of grayscale, color and ultrasound multimodal medical images show that the proposed method has advantages on retention of image salient features. Compared with other models, the fused image obtained by the proposed method has higher objective indexes with clearer edge contour, higher overall contrast, and no ringing effect and artifacts.

Designing and training of a dual CNN for image denoising

Deep convolutional neural networks (CNNs) for image denoising have recently attracted increasing research interest. However, plain networks cannot recover fine details for a complex task, such as real noisy images. In this paper, we propose a Dual denoising Network (DudeNet) to recover a clean image. Specifically, DudeNet consists of four modules: a feature extraction block, an enhancement block, a compression block, and a reconstruction block. The feature extraction block with a sparse mechanism extracts global and local features via two sub-networks. The enhancement block gathers and fuses the global and local features to provide complementary information for the latter network. The compression block refines the extracted information and compresses the network. Finally, the reconstruction block is utilized to reconstruct a denoised image. The DudeNet has the following advantages: (1) The dual networks with a sparse mechanism can extract complementary features to enhance the generalized ability of denoiser. (2) Fusing global and local features can extract salient features to recover fine details for complex noisy images. (3) A small-size filter is used to reduce the complexity of denoiser. Extensive experiments demonstrate the superiority of DudeNet over existing current state-of-the-art denoising methods. The code of DudeNet is accessible at https://github.com/hellloxiaotian/DudeNet.

A Logarithmic Function-Based Novel Representation Algorithm for Image Classification

Salient feature extraction is an important task in image classification and recognition. Although classification techniques focus on the bright part of an image, many pixels of the image are of similar saliency. To address the issue, this paper proposes the logarithmic function-based novel representation algorithm (LFNR) to apply a novel representation for each image. The original and novel representations were fused to improve the classification accuracy. Experimental results show that, thanks to the simultaneous use of original and novel representations, the test samples could be better classified. The classification algorithms coupled with the LFNR all witnessed lower error rates than the original algorithms. In particular, the collaboration representation-based classification coupled with the LFNR significantly outperformed the other sparse representation algorithms, such as homotopy, primal augmented Lagrangian method (PALM), and sparse reconstruction by separable approximation algorithm (SpaRSA). The no-parameter property of the LFNR is also noteworthy.

SIO: A Spatioimageomics Pipeline to Identify Prognostic Biomarkers Associated with the Ovarian Tumor Microenvironment.

Simple SummaryHigh-grade serous ovarian cancer (HGSC) caused more than 13,000 deaths annually in the United States. A critically important component that influences the HGSC patient survival is the tumor microenvironment. However, how different cells interact to influence HGSC patients’ survival remains largely unknown. To investigate this, we developed a pipeline that combines imaging mass cytometry (IMC), location-specific transcriptomics, and deep learning to identify the distribution of various stromal, tumor and immune cells as well as their spatial relationship. Our pipeline automatically and accurately segments cells and extracts salient cellular features to identify biomarkers, and multiple nearest-neighbor interactions among different cells that coordinate to influence overall survival rates in HGSC patients. In addition, we integrated IMC data with microdissected tumor and stromal transcriptomes to identify novel signaling networks. These results may lead to the discovery of novel survival rate-modulating mechanisms in HGSC patients.Stromal and immune cells in the tumor microenvironment (TME) have been shown to directly affect high-grade serous ovarian cancer (HGSC) malignant phenotypes, however, how these cells interact to influence HGSC patients’ survival remains largely unknown. To investigate the cell-cell communication in such a complex TME, we developed a SpatioImageOmics (SIO) pipeline that combines imaging mass cytometry (IMC), location-specific transcriptomics, and deep learning to identify the distribution of various stromal, tumor and immune cells as well as their spatial relationship in TME. The SIO pipeline automatically and accurately segments cells and extracts salient cellular features to identify biomarkers, and multiple nearest-neighbor interactions among tumor, immune, and stromal cells that coordinate to influence overall survival rates in HGSC patients. In addition, SIO integrates IMC data with microdissected tumor and stromal transcriptomes from the same patients to identify novel signaling networks, which would lead to the discovery of novel survival rate-modulating mechanisms in HGSC patients.

Multi-GAT: A Graphical Attention-Based Hierarchical Multimodal Representation Learning Approach for Human Activity Recognition

Recognizing human activities is one of the crucial capabilities that a robot needs to have to be useful around people. Although modern robots are equipped with various types of sensors, human activity recognition (HAR) still remains a challenging problem, particularly in the presence of noisy sensor data. In this work, we introduce a multimodal graphical attention-based HAR approach, called Multi-GAT, which hierarchically learns complementary multimodal features. We develop a multimodal mixture-of-experts model to disentangle and extract salient modality-specific features that enable feature interactions. Additionally, we introduce a novel message-passing based graphical attention approach to capture cross-modal relation for extracting complementary multimodal features. The experimental results on two multimodal human activity datasets suggest that Multi-GAT outperformed state-of-the-art HAR algorithms across all datasets and metrics tested. Finally, the experimental results with noisy sensor data indicate that Multi-GAT consistently outperforms all the evaluated baselines. The robust performance suggests that Multi-GAT can enable seamless human-robot collaboration in noisy human environments.

Twin-Incoherent Self-Expressive Locality-Adaptive Latent Dictionary Pair Learning for Classification.

The projective dictionary pair learning (DPL) model jointly seeks a synthesis dictionary and an analysis dictionary by extracting the block-diagonal coefficients with an incoherence-constrained analysis dictionary. However, DPL fails to discover the underlying subspaces and salient features at the same time, and it cannot encode the neighborhood information of the embedded coding coefficients, especially adaptively. In addition, although the data can be well reconstructed via the minimization of the reconstruction error, useful distinguishing salient feature information may be lost and incorporated into the noise term. In this article, we propose a novel self-expressive adaptive locality-preserving framework: twin-incoherent self-expressive latent DPL (SLatDPL). To capture the salient features from the samples, SLatDPL minimizes a latent reconstruction error by integrating the coefficient learning and salient feature extraction into a unified model, which can also be used to simultaneously discover the underlying subspaces and salient features. To make the coefficients block diagonal and ensure that the salient features are discriminative, our SLatDPL regularizes them by imposing a twin-incoherence constraint. Moreover, SLatDPL utilizes a self-expressive adaptive weighting strategy that uses normalized block-diagonal coefficients to preserve the locality of the codes and salient features. SLatDPL can use the class-specific reconstruction residual to handle new data directly. Extensive simulations on several public databases demonstrate the satisfactory performance of our SLatDPL compared with related methods.

Computational intelligence based secure three-party CBIR scheme for medical data for cloud-assisted healthcare applications

Medical images with various modalities have become an integral part of the diagnosis and treatment of several diseases. The medical practitioners often use previous case studies to deal with the current medical condition of any particular patient. In such circumstance, they need to securely access medical images of various cases which are generally stored in a network and are vulnerable to malicious attacks. To address these sensitive inadequacies, we have proposed computational intelligence based secure healthcare Content based Image Retrieval (CBIR) for medical image retrieval scheme through which any medical practitioner can retrieve the image in an encrypted domain in cloud environment. In this regard, hamming distance-based similarity matching is the only available technique that effectively handles the comparison between encrypted features. This technique requires binary features to perform similarity matching, and the performance of such features in image retrieval is poor. In this concern, we have suggested a salient component-based binary feature extraction approach to enhance retrieval accuracy. Initially, we have re-arranged the input image using the saliency map, principal texture direction, and entropy to place the salient components at the starting blocks. Subsequently, we have employed a block-level majority voting scheme on the salient blocks of the image to obtain local binary features. As a result, the final feature vector carries more features from the salient part of the image, which propitiously improves the retrieval accuracy. Later, we have encrypted the binary feature vector and performed image retrieval on cloud environment which involve Data Owner, Database Service Provider and Client over encrypted domain to full fill the security aspect. Finally, we have used medical as well as Corel image datasets to validate the retrieval performance accuracy of the proposed scheme. The experimental results obtained from real life datasets exhibit that the proposed method is secure and provides comparable retrieval accuracy concerning other related schemes in the domain.

TraMiner: Vision-Based Analysis of Locomotion Traces for Cognitive Assessment in Smart-Homes

The rapid increase in the senior population is posing serious challenges to national healthcare systems. Hence, innovative tools are needed to early detect health issues, including cognitive decline. Several clinical studies show that it is possible to identify cognitive impairment based on the locomotion patterns of the elderly. In this work, we investigate the use of sensor data and deep learning to recognize those patterns in instrumented smart-homes. In order to get rid of the noise introduced by indoor constraints and activity execution, we introduce novel visual feature extraction methods for locomotion data. Our solution relies on locomotion trace segmentation, image-based extraction of salient features from locomotion segments, and vision-based deep learning. We carried out extensive experiments with a large dataset acquired in a smart-home test bed from 153 seniors, including people with cognitive diseases. Results show that our system can accurately recognize the cognitive status of the senior, reaching a macro-F_1 score of 0.873 for the three categories that we target: cognitive health, mild cognitive impairment, and dementia. Moreover, an experimental comparison shows that our system outperforms state-of-the-art methods.

A Multi-FoV Viewport-Based Visual Saliency Model Using Adaptive Weighting Losses for 360$^\circ$ Images

360 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^\circ$</tex-math></inline-formula> media allows observers to explore the scene in all directions. The consequence is that the human visual attention is guided by not only the perceived area in the viewport but also the overall content in 360 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^\circ$</tex-math></inline-formula> . In this paper, we propose a method to estimate the 360 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^\circ$</tex-math></inline-formula> saliency map which extracts salient features from the entire 360 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^\circ$</tex-math></inline-formula> image in each viewport in three different Field of Views (FoVs). Our model is first pretrained with a large-scale 2D image dataset to enable the interpretation of semantic contents, then fine-tuned with a relative small 360 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$^\circ$</tex-math></inline-formula> image dataset. A novel weighting loss function attached with stretch weighted maps is introduced to adaptively weight the losses of three evaluation metrics and attenuate the impact of stretched regions in equirectangular projection during training process. Experimental results demonstrate that our model achieves better performance with the integration of three FoVs and its diverse viewport images. Results also show that the adaptive weighting losses and stretch weighted maps effectively enhance the evaluation scores compared to the fixed weighting losses solutions. Comparing to other state of the art models, our method surpasses them on three different datasets and ranks the top using 5 performance evaluation metrics on the Salient360! benchmark set. The code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/FannyChao/MV-SalGAN360</uri> .

STDFusionNet: An Infrared and Visible Image Fusion Network Based on Salient Target Detection

In this article, we propose an infrared and visible image fusion network based on the salient target detection, termed STDFusionNet, which can preserve the thermal targets in infrared images and the texture structures in visible images. First, a salient target mask is dedicated to annotating regions of the infrared image that humans or machines pay more attention to, so as to provide spatial guidance for the integration of different information. Second, we combine this salient target mask to design a specific loss function to guide the extraction and reconstruction of features. Specifically, the feature extraction network can selectively extract salient target features from infrared images and background texture features from visible images, while the feature reconstruction network can effectively fuse these features and reconstruct the desired results. It is worth noting that the salient target mask is only required in the training phase, which enables the proposed STDFusionNet to be an end-to-end model. In other words, our STDFusionNet can fulfill salient target detection and key information fusion in an implicit manner. Extensive qualitative and quantitative experiments demonstrate the superiority of our fusion algorithm over the current state of the arts, where our algorithm is much faster and the fusion results look like high-quality visible images with clear highlighted infrared targets. Moreover, the experimental results on the public datasets reveal that our algorithm can improve the entropy (EN), mutual information (MI), visual information fidelity (VIF), and spatial frequency (SF) metrics with about 1.25%, 22.65%, 4.3%, and 0.89% gains, respectively. Our code is publicly available at https://github.com/jiayi-ma/STDFusionNet .

Interleaved Deep Artifacts-Aware Attention Mechanism for Concrete Structural Defect Classification.

Automatic machine classification of concrete structural defects in images poses significant challenges because of multitude of problems arising from the surface texture, such as presence of stains, holes, colors, poster remains, graffiti, marking and painting, along with uncontrolled weather conditions and illuminations. In this paper, we propose an interleaved deep artifacts-aware attention mechanism (iDAAM) to classify multi-target multi-class and single-class defects from structural defect images. Our novel architecture is composed of interleaved fine-grained dense modules (FGDM) and concurrent dual attention modules (CDAM) to extract local discriminative features from concrete defect images. FGDM helps to aggregate multi-layer robust information with wide range of scales to describe visually-similar overlapping defects. On the other hand, CDAM selects multiple representations of highly localized overlapping defect features and encodes the crucial spatial regions from discriminative channels to address variations in texture, viewing angle, shape and size of overlapping defect classes. Within iDAAM, FGDM and CDAM are interleaved to extract salient discriminative features from multiple scales by constructing an end-to-end trainable network without any preprocessing steps, making the process fully automatic. Experimental results and extensive ablation studies on three publicly available large concrete defect datasets show that our proposed approach outperforms the current state-of-the-art methodologies.

Saliency Maps-Based Convolutional Neural Networks for Facial Expression Recognition

Facial expression recognition (FER) is one of the important research contents in affective computing. It plays a key role in many application fields of human life. As a most common expression feature extraction method, the convolutional neural network (CNN) has the following main limitation. Due to the fact that the CNN network lacks the visual attention guidance, when it gets expression information it brings background noises, resulting in the lower recognition accuracy. In order to simulate the attention mechanism in human visual system, a salient feature extraction model is proposed, including the dilated inception module, the Difference of Gaussian (DOG) module, and the multi-indicator saliency prediction module. This model can effectively reflect the key facial information through the increase of the receptive field, the acquisition of multiscale features, and the simulation of human vision. In addition, a novel FER method for one single person is proposed. With the prior knowledge of saliency maps and the multilayer deep features in the CNN network, the recognition accuracy is improved by obtaining more targeted and more complete deep expression information. The experimental results of saliency prediction, action unit (AU) detection, and smile intensity estimation on the CAT2000, the CK+, and the BP4D databases prove that the proposed method improves the FER performance and is more effective than the existing approaches.

Contactless Multi-biometric System Using Fingerprint and Palmprint Selfies

Due to the COVID-19 pandemic, automated contactless person identification based on the human hand has become very vital and an appealing biometric trait. Since, people are expected to cover their faces with masks, and advised avoiding touching surfaces. It is well-known that usually contact-based hand biometrics suffer from issues like deformation due to uneven distribution of pressure or improper placement on sensor, and hygienic concerns. Whereas, to mitigate such problems, contactless imaging is expected to collect the hand biometrics information without any deformation and leading to higher person recognition accuracy; besides maintaining hygienic and pandemic concerns. Towards this aim, in this paper, an effective multi-biometric scheme for person authentication based on contactless fingerprint and palmprint selfies has been proposed. In this study, for simplicity and efficiency, three local descriptors, i.e., local phase quantization (LPQ), local Ternary patterns (LTP), and binarized statistical image features (BSIF), have been employed to extract salient features from contactless fingerprint and palmprint selfies. The score level fusion based multi-biometric system developed in this work combines the matching scores using two different fusion techniques, i.e., transformation based-rules like triangular norms and classifier based-rules like SVM. Experimental results on two publicly available databases (i.e., PolyU contactless to contact-based fingerprint database and IIT-Delhi touchless palmprint dataset) show that the proposed contactless multi-biometric selfie system can easily outperform uni-biometrics.

Fusion-ConvBERT: Parallel Convolution and BERT Fusion for Speech Emotion Recognition.

Speech emotion recognition predicts the emotional state of a speaker based on the person’s speech. It brings an additional element for creating more natural human–computer interactions. Earlier studies on emotional recognition have been primarily based on handcrafted features and manual labels. With the advent of deep learning, there have been some efforts in applying the deep-network-based approach to the problem of emotion recognition. As deep learning automatically extracts salient features correlated to speaker emotion, it brings certain advantages over the handcrafted-feature-based methods. There are, however, some challenges in applying them to the emotion recognition problem, because data required for properly training deep networks are often lacking. Therefore, there is a need for a new deep-learning-based approach which can exploit available information from given speech signals to the maximum extent possible. Our proposed method, called “Fusion-ConvBERT”, is a parallel fusion model consisting of bidirectional encoder representations from transformers and convolutional neural networks. Extensive experiments were conducted on the proposed model using the EMO-DB and Interactive Emotional Dyadic Motion Capture Database emotion corpus, and it was shown that the proposed method outperformed state-of-the-art techniques in most of the test configurations.

Traffic System Anomaly Detection using Spatiotemporal Pattern Networks

Traffic dynamics in the urban interstate system are critical in terms of highway safety and mobility. This paper proposes a systematic data mining technique to detect traffic system-level anomalies in a batch-processing fashion. Built on the concepts of symbolic dynamics, a spatiotemporal pattern network (STPN) architecture is developed to capture the system characteristics. This novel spatiotemporal graphical modeling approach is shown to be able to extract salient time series features and discover spatial and temporal patterns for a traffic system. An information-theoretic metric is used to quantify the causal relationships between sub-systems. By comparing the structural similarity of the information-theoretic metrics of the STPNs learnt from each day, a day with anomalous system characteristics can be identified. A case study is conducted on an urban interstate in Iowa, USA, with 11 roadside radar sensors collecting 20-second resolution speed and volume data. After applying the proposed methods on one-month data (Feb. 2017), several system-level anomalies are detected. The potential causes that include inclement weather condition and non-recurring congestion are also verified to demonstrate the efficacies of the proposed technique. Compared to the traditional predefined performance measures for the traffic systems, the proposed framework has advantages in capturing spatiotemporal features in a fast and scalable manner.

Fusion and Visualization of Bridge Deck Nondestructive Evaluation Data via Machine Learning

To maintain infrastructure safety and integrity, nondestructive evaluation (NDE) technologies are often used for detection of subsurface defects and for holistic condition assessment of structures. While the rapid advances in data collection and the diversity of available sensing technologies provide new opportunities, the ability to efficiently process data and combine heterogeneous data sources to make robust decisions remains a challenge. Heterogeneous NDE measurements often conflict with one another and methods to visualize integrated results are often developed ad hoc. In this work, we present a framework to support fusion of multiple NDE techniques in order to improve both detection and quantification accuracy while also improving the visualization of NDE results. For data sources with waveform representations, the discrete wavelet transform (DWT) is used to extract salient features and facilitate fusion with scalar-valued NDE measurements. The description of a signal in terms of its salient features using a wavelet transform allows for capturing the significance of the original data, while suppressing measurement noise. The complete set of measurements is then fused using nonparametric machine learning so as to relax the need for Bayesian assumptions regarding statistical distributions. A novel visualization schema based on classifier confidence intervals is then employed to support holistic visualization and decision making. To validate the capabilities of the proposed methodology, an experimental prototype system was created and tested from NDE measurements of laboratory-scale bridge decks at Turner-Fairbank highway research center (TFHRC). The laboratory decks exhibit various types of artificial defects and several non-destructive tests have already been carried out by research center technicians to characterize the mechanical properties of the materials and the existing damages. The behaviors of several different data fusion approaches are compared here. Based on the presented analyses, it is demonstrated that fusion via support vector machines provided the most robust and consistent data fusion and defect detection capabilities.

A Deep-Learning Model for Automated Detection of Intense Midlatitude Convection Using Geostationary Satellite Images

AbstractIntense thunderstorms threaten life and property, impact aviation, and are a challenging forecast problem, particularly without precipitation-sensing radar data. Trained forecasters often look for features in geostationary satellite images such as rapid cloud growth, strong and persistent overshooting tops, U- or V-shaped patterns in storm-top temperature (and associated above-anvil cirrus plumes), thermal couplets, intricate texturing in cloud albedo (e.g., “bubbling” cloud tops), cloud-top divergence, spatial and temporal trends in lightning, and other nuances to identify intense thunderstorms. In this paper, a machine-learning algorithm was employed to automatically learn and extract salient features and patterns in geostationary satellite data for the prediction of intense convection. Namely, a convolutional neural network (CNN) was trained on 0.64-μm reflectance and 10.35-μm brightness temperature from the Advanced Baseline Imager (ABI) and flash-extent density (FED) from the Geostationary Lightning Mapper (GLM) on board GOES-16. Using a training dataset consisting of over 220 000 human-labeled satellite images, the CNN learned pertinent features that are known to be associated with intense convection and skillfully discriminated between intense and ordinary convection. The CNN also learned a more nuanced feature associated with intense convection—strong infrared brightness temperature gradients near cloud edges in the vicinity of the main updraft. A successive-permutation test ranked the most important predictors as follows: 1) ABI 10.35-μm brightness temperature, 2) ABI GLM flash-extent density, and 3) ABI 0.64-μm reflectance. The CNN model can provide forecasters with quantitative information that often foreshadows the occurrence of severe weather, day or night, over the full range of instrument-scan modes.