Discriminative Convolutional Neural Network Research Articles

Evaluating different combination methods to analyse ultrasound and shear wave elastography images automatically through discriminative convolutional neural network in breast cancer imaging

PurposeUltrasound (US) and Shear Wave Elastography (SWE) imaging are non-invasive methods used for breast lesion characterization. While US and SWE images provide both morphological information, SWE visualizes in addition the elasticity of tissue. In this study a Discriminative Convolutional Neural Network (DCNN) model is applied to US and SWE images and their combination to classify the breast lesions into malignant or benign cases. Furthermore, it is identified whether analysing only the region of the elastogram or including the surrounding B-mode image gives a superior performance.MethodsThe dataset used in this study consists of 746 images obtained from 207 patients comprising 486 malignant and 260 benign breast lesions. From each image the US and SWE image was extracted, once including only the region of the elastogram and once including also the surrounding B-mode image. These four datasets were applied individually to a DCNN to determine their predictive capability. Each the best US and SWE dataset were used to examine different combination methods with DCNN. The results were compared to the manual assessment by an expert radiologist.ResultsThe combination of US and SWE images with the surrounding B-mode image using two ensembled DCNN models achieved best results with an accuracy of 93.53 %, sensitivity of 94.42 %, specificity of 90.75 % and area under the curve (AUC) of 96.55 %.ConclusionThis study showed that using the whole US and SWE images through DCNN was superior to methods, in which only the region of elastogram was used. Combining breast cancer US and SWE images with two ensembled DCNN models in parallel improved the results. The accuracy, sensitivity and AUC of the best combination method were significantly superior to the results of using a single dataset through DCNN and to the results of the expert radiologist.

Two-Stage Approach of Hierarchical Deep Feature Representation and Fusion Frameworks for Brain Image Analysis

In recent decades, magnetic resonance (MR) brain images have initiated a wide range of image classification and segmentation methods. Feature representation is one of the essential aspects of medical image analysis. This paper proposes and investigates specific features that address the significance of high-level tasks with little annotation for medical images. Deep learning is a futuristic area of research in biomedical image analysis, in which the scope is moving us immediately to the goal of automating tasks in intelligent retrieval systems. This approach can incorporate many levels of feature representation to construct recognition of medical cells or images. We propose a novel approach based on the deep hierarchical features of two different convolutional neural network (CNNs) model choices to achieve competitive performance in the classification task. We explore feature representation through discriminative CNN models. The principal study of our proposed work is feature representations, feature-level fusion, and classification. Meanwhile, effective fusion frameworks were employed for brain MR image classification by using serial fusion and fusion operator strategies. The accuracy of the proposed technique is demonstrated using the Cancer Imaging Archive (TCIA) and Information eXtraction from Images (IXI) datasets. To the best of our knowledge, experiment results show that CNNs feature maps as input to the classifier and are superior to the original CNNs. The performance of the support vector machines (SVM) classifier has been used to evaluate in terms of training performance and classify subjects as either normal or abnormal.

Seismic Intelligent Deblending via Plug and Play Method With Blended CSGs Trained Deep CNN Gaussian Denoiser

Deblending can extract good quality seismic data from blended seismic data. Generally, the deblending methods can be categorized as model- and data-driven methods. The model-driven deblending methods usually suffer from a massive computational burden, while the data-driven ones need to implement a forward blending process to construct training samples for network training. To solve these issues, we develop a plug and play (PnP) method that integrates a trained convolutional neural network (CNN) Gaussian denoiser as the prior for seismic intelligent deblending. Specifically, we propose to use acquired blended common shot gathers (CSGs) as training dataset for the CNN Gaussian denoiser training to avoid constructing or collecting any extra data. According to the theory and dedicated designed experiments, this training mode can greatly improve network performance. Then, the trained CNN Gaussian denoiser is plugged into the alternating direction method of multiplier (ADMM) algorithm to solve the deblending problem. Furthermore, based on the <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${l}_{{\mathrm{2}}} $ </tex-math></inline-formula> norm data fidelity term and the special structure of the network architecture, the PnP-ADMM method converges to a fixed point. Experiments on synthetic and field data demonstrate that the presented PnP method with blended CSGs trained deep CNN Gaussian denoiser has superior deblending performances over the dictionary learning and discriminative deblending CNN methods.

MULTIMODAL BIOMETRIC IDENTIFICATION SYSTEM USING THE FUSION OF FINGERPRINT AND IRIS RECOGNITION WITH CNN APPROACH

Multimodal biometric systems are widely applied in many real-world applications because of its ability to accommodate variety of great limitations of unimodal biometric systems, including sensitivity to noise, population coverage, intra-class variability, nonuniversality, and vulnerability to spoofing. during this paper, an efficient and real-time multimodal biometric system is proposed supported building deep learning representations for images of both the correct and left irises of someone, and fusing the results obtained employing a ranking-level fusion method. The trained deep learning system proposed is named IrisConvNet whose architecture relies on a mix of Convolutional Neural Network (CNN) and Softmax classifier to extract discriminative features from the input image with none domain knowledge where the input image represents the localized iris region and so classify it into one amongst N classes. during this work, a discriminative CNN training scheme supported a mixture of back-propagation algorithm and mini-batch AdaGrad optimization method is proposed for weights updating and learning rate adaptation, respectively. additionally, other training strategies (e.g., dropout method, data augmentation) also are proposed so as to gauge different CNN architectures. The performance of the proposed system is tested on three public datasets collected under different conditions: SDUMLA-HMT, CASIA-IrisV3 Interval and IITD iris database

Enriched and discriminative convolutional neural network features for pedestrian re‐identification and trajectory modeling

AbstractUnderstanding pedestrian flow patterns in urban areas could support the decision‐making for infrastructure planning. By incorporating computer vision techniques into surveillance video processing, human walking trajectories in a wide area could be identified by pedestrian re‐identification (ReID) across multiple cameras. Recent ReID methods mostly use convolutional neural networks equipped with deep learning techniques to extract discriminative human features from images for identity matching. However, they still suffer from realistic challenges such as occlusion and appearance variation. This paper develops a ReID‐based framework for pedestrian trajectory recognition across multiple cameras. Specifically, a generic approach of explainable model design is presented, which intuitively analyzes existing baseline models based on feature visualization. Hence, a new model named OSNet + BDB is developed that extracts discriminative‐and‐distributed features. Additionally, an incremental feature aggregation strategy is designed for more robust identity matching. Our ReID method notably outperforms its baselines by 4% identification F1 accuracy in public benchmarks. Practically, pedestrian flow statistics in a real building are extracted for behavioral modeling. Simulations of several what‐if layouts are then conducted for facility performance evaluation.

Abstract P05: EpistoNet: An ensemble of deep convolutional neural networks using mixture of discriminative experts for detecting COVID-19 on chest X-ray images

Abstract Introduction: The Coronavirus has spread across the globe and infected millions of people, having devastating effect on the global public health and economies. A fast diagnostic system should be implemented to mitigate the impact of the virus and save lives. In this study, we propose a decision tree-based ensemble model using two mixtures of discriminative experts (MoE) to classify COVID-19 and non-COVID-19 lung infections on chest X-ray images. The Epistocracy algorithm, a hyper-heuristic evolutionary method, is employed to optimize the neural networks used in this work. Using this approach can help detect COVID-19 cases and accelerate treatment of those who need it the most. Data: we collected 2,500 chest X-ray images from Henry Ford Health System consisting of 1,250 Covid images and 1,250 non-Covid images. The input images have been cropped and resized to 224 by 224 pixels. Out of 2,500 images, we left out 500 images containing 250 Covid and 250 non-Covid for testing. The rest, 2,000 images, were used 80% for training and 20% for validation. Methods and Results: To improve the accuracy of the proposed model, first we divided our 2,000 images into 5 different clusters using K-Means clustering algorithm with VGG16 feature extractor to help build strong discriminative expert models to be used in our proposed classifier. We trained VGG16, VGG19, InceptionV3, InceptionResNetV2, MobileNetV2, EfficientNetB7, Xception, and DenseNet201 to classify each cluster into Covid and non-Covid cases. The best result was obtained from VGG16 as a base model with a deep neural network as a head model optimized by Epistocracy algorithm. Then we built a mixture of transfer learning-based experts consisting of 5 different VGG16 models supervised by InceptionV3 as a gating network. Finally, we built a decision tree-based ensemble model to determine the classification of the data using two different MoEs with highest accuracies. As a result, for initial clusters c1, c2, c3, c4, and c5 we obtained validation accuracy of 92.50%, 86.30%, 86.51%, 85.34%, and 93.62% respectively. The first MoE had 93.75% accuracy on validation, and the second MoE had 94.25%. The final ensemble model on average obtained 94% accuracy on the testing dataset. More specifically, we got 96% accuracy on Covid images and 92% accuracy on non-Covid. Conclusion: we showed that an ensemble model consisting of two mixtures of cluster-based discriminative convolutional neural network experts can be used to detect Covid from non-Covid with high accuracy, and Epistocracy algorithm can be effectively used to optimize the hyper-parameters of the proposed models. Citation Format: Seyed Ziae Mousavi Mojab, Seyedmohammad Shams, Farshad Fotouhi, Hamid Soltanian-Zadeh. EpistoNet: An ensemble of deep convolutional neural networks using mixture of discriminative experts for detecting COVID-19 on chest X-ray images [abstract]. In: Proceedings of the AACR Virtual Meeting: COVID-19 and Cancer; 2021 Feb 3-5. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(6_Suppl):Abstract nr P05.

Leaf image-based classification of some common bean cultivars using discriminative convolutional neural networks

In recent years, many efforts have been made to apply image processing techniques for plant leaf identification. However, categorizing leaf images at the cultivar/variety level, because of the very low inter-class variability, is still a challenging task. In this research, we propose an automatic discriminative method based on convolutional neural networks (CNNs) for classifying 12 different cultivars of common beans that belong to three various species. We show that employing advanced loss functions, such as Additive Angular Margin Loss and Large Margin Cosine Loss, instead of the standard softmax loss function for the classification can yield better discrimination between classes and thereby mitigate the problem of low inter-class variability. The method was evaluated by classifying species (level I), cultivars from the same species (level II), and cultivars from different species (level III), based on images from the leaf foreside and backside. The results indicate that the performance of the classification algorithm on the leaf backside image dataset is superior. The maximum mean classification accuracies of 95.86, 91.37 and 86.87% were obtained at the levels I, II and III, respectively. The proposed method outperforms the previous relevant works and provides a reliable approach for plant cultivars identification.

Hyperspectral image classification via discriminative convolutional neural network with an improved triplet loss

Hyper-Spectral Image (HSI) classification is an important task because of its wide range of applications. With the remarkable success from the Convolutional Neural Network (CNN), the performance of HSI classification has been significantly improved. However, two main challenges remained. One is that the samples of HSI have dramatic intra-class diversity and inter-class similarity, and the conventional cross-entropy loss is not good enough to learn discriminative features. The other is that the number of the training samples is so limited that the network is easy to overfit. To address the first challenge, we develop an improved triplet loss in order to make samples from the same class close to each other and make samples from different classes further apart. The proposed loss function considers all the possible positive pairs and negative pairs in a training batch, filters many trivial pairs, and prevents the impact of the outliers at the same time. To deal with the second challenge, we design an appropriate network architecture with less learnable parameters. We train the designed network based on the proposed loss with randomly initialized network weights using only hundreds of training samples, and attain quite good results. The experimental results show that the proposed method significantly surpasses other state-of-the-art methods, especially with less training samples. Furthermore, being less complex, the training process only takes a few minutes on a single GPU, which is faster than other state-of-the-art CNN-based methods.

Deep learning for non-rigid 3D shape classification based on informative images

In order to enhance the discernment of features in view-based 3D shape recognition, we propose a joint convolutional neural network (CNN) learning model based on informative images. It learns deep features from intrinsic feature images and extrinsic 2D views, and generates a synthetic feature vector via weighted aggregation and refinement process, which has achieved remarkable improvement in non-rigid 3D shape classification. Our joint CNNs model contains three parts: the first part is the geometry-based feature generation unit. We provide a discriminative BoF (bag of features) image descriptor and construct CNN framework to learn the geometric features of the model. The second part is the view-based feature generation unit. We establish a parallel CNN to extract spatial features from optimized 2D views. The third part is a score generation and refinement unit, which automatically learns the weighted scores of geometric features and spatial features. Finally, the aggregated feature is refined in a CNN framework and serves as an informative shape descriptor for recognition task. The experimental results demonstrate that our deep features have the strong discerning ability. Thus, better performance and robustness can be obtained compared to state-of-the-art methods.

A Discriminative Convolutional Neural Network with Context-aware Attention

Feature representation and feature extraction are two crucial procedures in text mining. Convolutional Neural Networks (CNN) have shown overwhelming success for text-mining tasks, since they are capable of efficiently extracting n -gram features from source data. However, vanilla CNN has its own weaknesses on feature representation and feature extraction. A certain amount of filters in CNN are inevitably duplicate and thus hinder to discriminatively represent a given text. In addition, most existing CNN models extract features in a fixed way (i.e., max pooling) that either limit the CNN to local optimum nor without considering the relation between all features, thereby unable to learn a contextual n -gram features adaptively. In this article, we propose a discriminative CNN with context-aware attention to solve the challenges of vanilla CNN. Specifically, our model mainly encourages discrimination across different filters via maximizing their earth mover distances and estimates the salience of feature candidates by considering the relation between context features. We validate carefully our findings against baselines on five benchmark datasets of classification and two datasets of summarization. The results of the experiments verify the competitive performance of our proposed model.

An Adversarial Discriminative Convolutional Neural Network for Cross-Project Defect Prediction

Cross-project defect prediction (CPDP) is a promising approach to help to allocate testing efforts efficiently and guarantee software reliability in the early software lifecycle. A CPDP method usually trains a software defect classifier based on labeled data sets. Then the trained classifier can predict new projects without labeled data. Most previous CPDP techniques focused on manually designing handcrafted features. However, these handcrafted features ignore the programs’ semantic information. Moreover, some other existing defect prediction approaches learned semantic features from source code to build classifiers directly. However, they did not consider the distribution divergence between source and target projects. To address these limitations, we put forward a new method called Adversarial Discriminative Convolutional Neural Network (ADCNN). It can extract the transferable semantic features from source code for CPDP tasks. Specifically, we first parse source files into token vectors and then map them to integer vectors via word embedding. Second, we combine adversarial learning with discriminative feature learning to train the ADCNN model. The key of the ADCNN model is to learn the discriminative mapping of the target project to the source feature space by deceiving a domain discriminator. A domain discriminator tries to distinguish the target project files from the source project files. Finally, we use the extracted transferable semantic features to build a classifier for CPDP tasks. We evaluate our method on ten benchmark projects in terms of F-measure, AUC, and PofB20 (an effort-aware evaluation metric). The experimental results demonstrate that our ADCNN method performs better compared with other related CPDP methods.

Discriminative Convolutional Neural Network for Image Quality Assessment with Fixed Convolution Filters

Discriminative Convolutional Neural Network for Image Quality Assessment with Fixed Convolution Filters

A Flexible Deep CNN Framework for Image Restoration

Image restoration is a long-standing problem in image processing and low-level computer vision. Recently, discriminative convolutional neural network (CNN)-based approaches have attracted considerable attention due to their superior performance. However, most of these frameworks are designed for one specific image restoration task; hence, they seldom show high performance on other image restoration tasks. To address this issue, we propose a flexible deep CNN framework that exploits the frequency characteristics of different types of artifacts. Hence, the same approach can be employed for a variety of image restoration tasks by adjusting the architecture. For reducing the artifacts with similar frequency characteristics, a quality enhancement network that adopts residual and recursive learning is proposed. Residual learning is utilized to speed up the training process and boost the performance; recursive learning is adopted to significantly reduce the number of training parameters as well as boost the performance. Moreover, lateral connections transmit the extracted features between different frequency streams via multiple paths. One aggregation network combines the outputs of these streams to further enhance the restored images. We demonstrate the capabilities of the proposed framework with three representative applications: image compression artifacts reduction (CAR), image denoising, and single image super-resolution (SISR). Extensive experiments confirm that the proposed framework outperforms the state-of-the-art approaches on benchmark datasets for these applications.

Verification of Biometric Traits using Deep Learning

Multimodal biometric systems have been widely applied in many real-world applications due to its ability to deal with a number of significant limitations of unimodal biometric systems including non-universality, noise, population coverage, vulnerability and intra-class variability for verification, authentication and identification of an individual. In this paper, the impact of deep learning in the field of biometrics is investigated where supervised learning is primarily involved in identifying biometric traits using Graphical User Interface. The trained deep learning system proposed is called MultiTraitConvNet whose architecture is based on a combination of Convolutional Neural Network (CNN) and Softmax classifier to extract discriminative features from the input image without prior domain knowledge and classify the class of the biometric trait. A discriminative CNN training scheme is based on a combination of back-propagation algorithm and mini-batch Adam optimization method that is used for weights updating and learning rate adaption, respectively. To evaluate various CNN architecture data augmentation and dropout method training techniques have been used. A centralized database of different biometric trait images has been created using data sets of CASIA1 for iris trait, google 11K hand for palmprint trait, for face UTKFace is used. The accuracy of the proposed system is found to be 100%. Python library Keras is used to develop CNN model and TKinter is used to create GUI of the proposed system.

Unsupervised Deep Noise Modeling for Hyperspectral Image Change Detection

Hyperspectral image (HSI) change detection plays an important role in remote sensing applications, and considerable research has been done focused on improving change detection performance. However, the high dimension of hyperspectral data makes it hard to extract discriminative features for hyperspectral processing tasks. Though deep convolutional neural networks (CNN) have superior capability in high-level semantic feature learning, it is difficult to employ CNN for change detection tasks. As a ground truth map is usually used for the evaluation of change detection algorithms, it cannot be directly used for supervised learning. In order to better extract discriminative CNN features, a novel noise modeling-based unsupervised fully convolutional network (FCN) framework is presented for HSI change detection in this paper. Specifically, the proposed method utilizes the change detection maps of existing unsupervised change detection methods to train the deep CNN, and then removes the noise during the end-to-end training process. The main contributions of this paper are threefold: (1) A new end-to-end FCN-based deep network architecture for HSI change detection is presented with powerful learning features; (2) An unsupervised noise modeling method is introduced for the robust training of the proposed deep network; (3) Experimental results on three datasets confirm the effectiveness of the proposed method.

Adversarial Deep Domain Adaptation for Multi-Band SAR Images Classification

Deep convolutional neural networks (CNNs) have made a breakthrough on supervised SAR images classification. However, SAR imaging is considerably affected by the frequency band. That means a neural network trained on a SAR image set of one band is not suitable for the classification of another band images. As manually labeling the training samples of each band is always time-consuming, we propose an unsupervised multi-level domain adaptation method based on adversarial learning to solve the problem of multi-band SAR images classification. First, we train a discriminative CNN using samples of one frequency band data set that contains labels to map the data to a latent feature space. Then, we adjust the trained CNN to map the unlabeled samples of another frequency band data set to the same feature space through alternately optimizing two adversarial loss functions. Thus, the features of these two band images are fused and can be classified by the same classifier. We checked the performance of our method using both simulated data and measured data. Our method made a breakthrough in the classification of multi-band images with accuracies of 99% on both data sets. The results are even very close to the supervised CNN trained using a large number of labeled samples.

Learning Rotation-Invariant and Fisher Discriminative Convolutional Neural Networks for Object Detection

The performance of object detection has recently been significantly improved due to the powerful features learnt through convolutional neural networks (CNNs). Despite the remarkable success, there are still several major challenges in object detection, including object rotation, within-class diversity, and between-class similarity, which generally degenerate object detection performance. To address these issues, we build up the existing state-of-the-art object detection systems and propose a simple but effective method to train rotation-invariant and Fisher discriminative CNN models to further boost object detection performance. This is achieved by optimizing a new objective function that explicitly imposes a rotation-invariant regularizer and a Fisher discrimination regularizer on the CNN features. Specifically, the first regularizer enforces the CNN feature representations of the training samples before and after rotation to be mapped closely to each other in order to achieve rotation-invariance. The second regularizer constrains the CNN features to have small within-class scatter but large between-class separation. We implement our proposed method under four popular object detection frameworks, including region-CNN (R-CNN), Fast R- CNN, Faster R- CNN, and R- FCN. In the experiments, we comprehensively evaluate the proposed method on the PASCAL VOC 2007 and 2012 data sets and a publicly available aerial image data set. Our proposed methods outperform the existing baseline methods and achieve the state-of-the-art results.

Generative Adversarial Networks for Hyperspectral Image Classification

A generative adversarial network (GAN) usually contains a generative network and a discriminative network in competition with each other. The GAN has shown its capability in a variety of applications. In this paper, the usefulness and effectiveness of GAN for classification of hyperspectral images (HSIs) are explored for the first time. In the proposed GAN, a convolutional neural network (CNN) is designed to discriminate the inputs and another CNN is used to generate so-called fake inputs. The aforementioned CNNs are trained together: the generative CNN tries to generate fake inputs that are as real as possible, and the discriminative CNN tries to classify the real and fake inputs. This kind of adversarial training improves the generalization capability of the discriminative CNN, which is really important when the training samples are limited. Specifically, we propose two schemes: 1) a well-designed 1D-GAN as a spectral classifier and 2) a robust 3D-GAN as a spectral–spatial classifier. Furthermore, the generated adversarial samples are used with real training samples to fine-tune the discriminative CNN, which improves the final classification performance. The proposed classifiers are carried out on three widely used hyperspectral data sets: Salinas, Indiana Pines, and Kennedy Space Center. The obtained results reveal that the proposed models provide competitive results compared to the state-of-the-art methods. In addition, the proposed GANs open new opportunities in the remote sensing community for the challenging task of HSI classification and also reveal the huge potential of GAN-based methods for the analysis of such complex and inherently nonlinear data.

When Deep Learning Meets Metric Learning: Remote Sensing Image Scene Classification via Learning Discriminative CNNs

Remote sensing image scene classification is an active and challenging task driven by many applications. More recently, with the advances of deep learning models especially convolutional neural networks (CNNs), the performance of remote sensing image scene classification has been significantly improved due to the powerful feature representations learnt through CNNs. Although great success has been obtained so far, the problems of within-class diversity and between-class similarity are still two big challenges. To address these problems, in this paper, we propose a simple but effective method to learn discriminative CNNs (D-CNNs) to boost the performance of remote sensing image scene classification. Different from the traditional CNN models that minimize only the cross entropy loss, our proposed D-CNN models are trained by optimizing a new discriminative objective function. To this end, apart from minimizing the classification error, we also explicitly impose a metric learning regularization term on the CNN features. The metric learning regularization enforces the D-CNN models to be more discriminative so that, in the new D-CNN feature spaces, the images from the same scene class are mapped closely to each other and the images of different classes are mapped as farther apart as possible. In the experiments, we comprehensively evaluate the proposed method on three publicly available benchmark data sets using three off-the-shelf CNN models. Experimental results demonstrate that our proposed D-CNN methods outperform the existing baseline methods and achieve state-of-the-art results on all three data sets.

More discriminative convolutional neural network with inter-class constraint for classification

Recently, convolutional neural network (CNN) has achieved impressive results in object classification tasks. Through various modifications and careful design of the inner structure of CNN, its performance has already become human-competitive, according to recent reports. To some extent, the testing accuracy depends largely on the decision boundaries produced by classifiers, which classify different objects into specific feature spaces. Hence, the relationships among samples in the feature space are critical. However, the softmax loss function that is used in most CNN models, does not directly contain the relationship information. In this paper, we propose a novel loss function, named inter-class constraint loss function, that maximizes the distance between different classes. Together with softmax loss, we can obtain larger inter-class distances and smaller intra-class distances in CNN, thus significantly improving the accuracy in classification. We achieve substantial improvements for the SVHN, CIFAR-10 and CIFAR-100 datasets using our proposed loss function.