Convolutional Neural Network Training Research Articles

Person re-identification accuracy improvement by training a CNN with the new large joint dataset and re-rank

The paper is aimed to improve person re-identification accuracy in distributed video surveillance systems based on constructing a large joint image dataset of people for training convolutional neural networks (CNN). For this aim, an analysis of existing datasets is provided. Then, a new large joint dataset for person re-identification task is constructed that includes the existing public datasets CUHK02, CUHK03, Market, Duke, MSMT17 and PolReID. Testing for re-identification is performed for such frequently cited CNNs as ResNet-50, DenseNet121 and PCB. Re-identification accuracy is evaluated by using the main metrics Rank, mAP and mINP. The use of the new large joint dataset makes it possible to improve Rank1 mAP, mINP on all test sets. Re-ranking is used to further increase the re-identification accuracy. Presented results confirm the effectiveness of the proposed approach.

BCM3D 2.0: accurate segmentation of single bacterial cells in dense biofilms using computationally generated intermediate image representations

Accurate detection and segmentation of single cells in three-dimensional (3D) fluorescence time-lapse images is essential for observing individual cell behaviors in large bacterial communities called biofilms. Recent progress in machine-learning-based image analysis is providing this capability with ever-increasing accuracy. Leveraging the capabilities of deep convolutional neural networks (CNNs), we recently developed bacterial cell morphometry in 3D (BCM3D), an integrated image analysis pipeline that combines deep learning with conventional image analysis to detect and segment single biofilm-dwelling cells in 3D fluorescence images. While the first release of BCM3D (BCM3D 1.0) achieved state-of-the-art 3D bacterial cell segmentation accuracies, low signal-to-background ratios (SBRs) and images of very dense biofilms remained challenging. Here, we present BCM3D 2.0 to address this challenge. BCM3D 2.0 is entirely complementary to the approach utilized in BCM3D 1.0. Instead of training CNNs to perform voxel classification, we trained CNNs to translate 3D fluorescence images into intermediate 3D image representations that are, when combined appropriately, more amenable to conventional mathematical image processing than a single experimental image. Using this approach, improved segmentation results are obtained even for very low SBRs and/or high cell density biofilm images. The improved cell segmentation accuracies in turn enable improved accuracies of tracking individual cells through 3D space and time. This capability opens the door to investigating time-dependent phenomena in bacterial biofilms at the cellular level.

Data Augmentation for Small Sample Iris Image Based on a Modified Sparrow Search Algorithm

Training convolutional neural networks (CNN) often require a large amount of data. However, for some biometric data, such as fingerprints and iris, it is often difficult to obtain a large amount of data due to privacy issues. Therefore, training the CNN model often suffers from specific problems, such as overfitting, low accuracy, poor generalization ability, etc. To solve them, we propose a novel image augmentation algorithm for small sample iris image in this article. It is based on a modified sparrow search algorithm (SSA) called chaotic Pareto sparrow search algorithm (CPSSA), combined with contrast limited adaptive histogram equalization (CLAHE). The CPSSA is used to search for a group of clipping limit values. Then a set of iris images that satisfies the constraint condition is produced by CLAHE. In the fitness function, cosine similarity is used to ensure that the generated images are in the same class as the original one. We select 200 categories of iris images from the CASIA-Iris-Thousand dataset and test the proposed augmentation method on four CNN models. The experimental results show that, compared with the some standard image augmentation methods such as flipping, mirroring and clipping, the accuracy and Equal Error Rate (EER)of the proposed method have been significantly improved. The accuracy and EER of the CNN models with the best recognition performance can reach 95.5 and 0.6809 respectively. This fully shows that the data augmentation method proposed in this paper is effective and quite simple to implement.

Damage detection in slab structures based on two-dimensional curvature mode shape method and Faster R-CNN

This paper proposes a novel method based on the two-dimensional (2D) curvature mode shape method, Convolutional Neural Networks (CNN), and Faster Region-based Convolutional Neural Networks (faster R-CNN) for detecting damage in slab structures. The 2D curvature mode shapes could be measured directly or calculated from the measured mode shapes using the central difference method. The damage indicator is defined as the absolute differences between the 2D curvature mode shapes of the damaged and intact slabs. The contour plot is chosen to convert the damage indicators into images. Four hundred damage scenarios are created using a Finite Element (FE) model of the slab. Images created from those damage scenarios are then used to train CNN and faster R-CNN. Four damage types are considered in this research, namely a single small hole, a single big hole, two small holes, and two large holes. After training, CNN can predict the damage types and faster R-CNN can predict the bounding boxes around the damaged areas. A test sample set is created to test the performance of the proposed method. The effect of noise in the mode shape data is considered. Results show that the classification accuracy for damage type is high. The overlap ratios between the predicted bounding boxes and the real damaged areas are more than 40% for 80% of tested scenarios. Furthermore, the low influence of noise on the predicted results is investigated. The proposed method is robust and has great potential for application to real structures.

SegMix: Co-occurrence Driven Mixup for Semantic Segmentation and Adversarial Robustness

In this paper, we present a strategy for training convolutional neural networks to effectively resolve interference arising from competing hypotheses relating to inter-categorical information throughout the network. In this work, this is accomplished for the task of dense image labelling by blending images based on (i) categorical clustering or (ii) the co-occurrence likelihood of categories. We then train a source separation network which simultaneously segments and separates the blended images. Subsequent feature denoising to suppress noisy activations reveals additional desirable properties and high degrees of successful predictions. Through this process, we reveal a general mechanism, distinct from any prior methods, for boosting the performance of the base segmentation and salient object detection network while simultaneously increasing robustness to adversarial attacks.

Convolutional neural network -based phantom image scoring for mammography quality control

BackgroundVisual evaluation of phantom images is an important, but time-consuming part of mammography quality control (QC). Consistent scoring of phantom images over the device’s lifetime is highly desirable. Recently, convolutional neural networks (CNNs) have been applied to a wide range of image classification problems, performing with a high accuracy. The purpose of this study was to automate mammography QC phantom scoring task by training CNN models to mimic a human reviewer.MethodsEight CNN variations consisting of three to ten convolutional layers were trained for detecting targets (fibres, microcalcifications and masses) in American College of Radiology (ACR) accreditation phantom images and the results were compared with human scoring. Regular and artificially degraded/improved QC phantom images from eight mammography devices were visually evaluated by one reviewer. These images were used in training the CNN models. A separate test set consisted of daily QC images from the eight devices and separately acquired images with varying dose levels. These were scored by four reviewers and considered the ground truth for CNN performance testing.ResultsAlthough hyper-parameter search space was limited, an optimal network depth after which additional layers resulted in decreased accuracy was identified. The highest scoring accuracy (95%) was achieved with the CNN consisting of six convolutional layers. The highest deviation between the CNN and the reviewers was found at lowest dose levels. No significant difference emerged between the visual reviews and CNN results except in case of smallest masses.ConclusionA CNN-based automatic mammography QC phantom scoring system can score phantom images in a good agreement with human reviewers, and can therefore be of benefit in mammography QC.

A Review on Real-Time Traffic Sign Recognition with Voice Warnings

Road signs are essential for providing information to drivers. Understanding road signs are essential for ensuring traffic safety because doing so can stop 4484 accidents. The identification of traffic signs has been the focus of research in recent decades. Accurate real-time recognition is the cornerstone of a robust but underdeveloped traffic sign recognition system. This study provides drivers with real-time voice-advice traffic sign recognition technology. This system is composed of two subsystems. Using a trained convolutional neural network, the first recognizes and detects traffic signs (CNN). When the system notices a particular traffic sign, the text-to-speech engine is employed to play a voice message to the driver. An efficient- CNN model is built on the reference data set using deep learning methods for search and real-time search. This system's advantage is that it recognizes traffic signs and guides the car even if the driver overlooks, ignores, or doesn't understand them. Say. These technologies are also necessary for the development of autonomous vehicles.

Research on Fault Diagnosis Method Based on Improved CNN

Traditional fault diagnosis methods require complex signal processing and expert experience, and the accuracy of fault identification is low. To solve these problems, a fault diagnosis method based on an improved convolutional neural network (CNN) is proposed. Based on the traditional CNN model, a variety of convolution stride modes were added to extract features of different scales of signals and expand the feature dimension. Firstly, the vibration signals were collected and grouped. Then, the data were divided into a training set and a test set and input into improved CNN for feature extraction and model training to realize fault identification. The proposed model achieved a classification accuracy of 99.3% when testing the vibration data of the armored vehicle. Finally, the proposed model was used to classify different fault types of planetary gearboxes. The gradient-weighted class activation mapping (Grad-CAM) method was used to visualize the classification weight of samples. The results showed that the classification accuracy reaches 98% under various working conditions of the planetary gearbox.

LDA-CNN: Linear Discriminant Analysis Convolution Neural Network for Periocular Recognition in the Wild

Due to the COVID-19 pandemic, the necessity for a contactless biometric system able to recognize masked faces drew attention to the periocular region as a valuable biometric trait. However, periocular recognition remains challenging for deployments in the wild or in unconstrained environments where images are captured under non-ideal conditions with large variations in illumination, occlusion, pose, and resolution. These variations increase within-class variability and between-class similarity, which degrades the discriminative power of the features extracted from the periocular trait. Despite the remarkable success of convolutional neural network (CNN) training, CNN requires a huge volume of data, which is not available for periocular recognition. In addition, the focus is on reducing the loss between the actual class and the predicted class but not on learning the discriminative features. To address these problems, in this paper we used a pre-trained CNN model as a backbone and introduced an effective deep CNN periocular recognition model, called linear discriminant analysis CNN (LDA-CNN), where an LDA layer was incorporated after the last convolution layer of the backbone model. The LDA layer enforced the model to learn features so that the within-class variation was small, and the between-class separation was large. Finally, a new fully connected (FC) layer with softmax activation was added after the LDA layer, and it was fine-tuned in an end-to-end manner. Our proposed model was extensively evaluated using the following four benchmark unconstrained periocular datasets: UFPR, UBIRIS.v2, VISOB, and UBIPr. The experimental results indicated that LDA-CNN outperformed the state-of-the-art methods for periocular recognition in unconstrained environments. To interpret the performance, we visualized the discriminative power of the features extracted from different layers of the LDA-CNN model using the t-distributed Stochastic Neighboring Embedding (t-SNE) visualization technique. Moreover, we conducted cross-condition experiments (cross-light, cross-sensor, cross-eye, cross-pose, and cross-database) that proved the ability of the proposed model to generalize well to different unconstrained conditions.

Application of deep learning neural networks for the analysis of fluid-particle dynamics in fibrous filters

A novel hybrid data-driven framework was introduced in this study for the prediction of fluid-particle dynamics of submicron particles under hydrodynamic and Brownian forces in fibrous filters. The distribution of flow parameters was predicted by a high-resolution flow surrogate model based on a deep convolutional neural network (CNN) coupled with a Lagrangian particle tracking method based on the discrete phase model (DPM). The CNN-DPM model was applied to a group of several nonuniformly distributed fibers to calculate filtration efficiency. For CNN training, ground-truth data were obtained using a two-dimensional computational fluid dynamics (CFD) model for laminar incompressible flow. A simulation speedup of three orders of magnitude was obtained by CNN for predicting the velocity components and static pressure distribution within an error range of ±10%. The generalizability of CNN model was also confirmed for overlapped irregular fiber shapes and porosities out of the range of the training dataset with deviations in a range of ±20% in comparison with CFD results. The collection efficiency by CNN-DPM for small particle diameters with dp≤0.3μm was overestimated by approximately 3 % while agreeing well with CFD-DPM for larger particles. The errors of flow velocity predictions by CNN near the fibers added more randomness to the particle motion which was combined with the random Brownian motion of particles, consequently increasing the probability of the small particle hit number and filtration efficiency. The overall acceptable accuracy of the CNN-DPM was confirmed, which is suitable for fibrous filter design and optimization purposes that require thousands of simulations.

Detection and Prevention of DDoS Attacks on the IoT

The Internet of Things (IoT) system has been a hot topic in recent years. Its operation is a system that stores data in data storage and is completed by the exchange of network information about things. Therefore, the security of information between network transmissions is very important. In recent years, the most likely cause of information security problems has been a distributed denial of service (DDoS) attack. In this paper, we proposed an autonomous defense system that combines edge computing with a two-dimensional convolutional neural network (CNN) to recognize whether the data server in IoT suffers from DDoS attacks and identify the attack mode. The accuracy of trained two-dimensional CNN is up to 99.5% and 99.8% for packet traffic and packet features training, respectively. A field experiment’s results show that the data server in the proposed system can effectively distinguish the difference between the DDoS attacks and the normal transmission to reduce the impact of DDoS attacks on the IoT data storage while it is under attack.

1-ADM-CNN: A Lightweight In-field Compression Method for Seismic Data

Large-scale seismic acquisition, versatility, flexibility, automation, and scalability are the objectives of future oil and gas exploration technology. An example of emerging technology for seismic monitoring is distributed acoustic sensing (DAS). The significant amount of data produced by DAS is a challenge that necessitates the development of new technologies for its efficient handling and processing. A typical seismic survey, on the one hand, can generate hundreds of terabytes of raw seismic data per day. The demand for wireless seismic data transmission, on the other hand, remains enormous. The massive amount of data transmission from geophones to the on-site data collection center and its storage poses significant challenges. A lightweight compression procedure is required in order to reduce the data traffic and the storage size at the data center without putting an extra burden on a geophone. In this brief, an efficient implementation of a 1D convolutional neural network (CNN) together with 1-bit adaptive delta modulation is presented for in-field seismic data compression. It is worth mentioning here that the training of CNN is done offline on synthetic data set and hence, the proposed approach has potential for real-time implementation. Furthermore, no assumption on the underlying statistics of noise or the seismic signal is imposed and consequently, the proposed method is suitable for a wide range of seismic data. Furthermore, the proposed method works in the time-domain, unlike existing transform-domain methods, making it suitable for quick diagnosis of bad traces at the data center. Simulation results with real data set reveal that the proposed approach achieves a signal-to-noise ratio (SNR) of approximately 30 dB with a compression gain of 35: 1. Finally, significant superiority in terms of compression gain and reconstruction quality is demonstrated when compared to the existing methods.

Exploring the Potential of Low-Bit Training of Convolutional Neural Networks

Convolutional neural networks (CNNs) have been widely used in many tasks, but training CNNs is time consuming and energy hungry. Using the low-bit integer format has been proved promising for speeding up and improving the energy efficiency of CNN inference, while CNN training can hardly benefit from such a technique because of the following challenges: 1) the integer data format cannot meet the requirements of the data dynamic range in training, resulting in the accuracy drop; 2) the floating-point data format keeps sizeable dynamic range with much more exponent bits, thus using it results in higher accumulation power than using the integer data format; and 3) there are some specially designed data formats (e.g., with group-wise scaling) that have the potential to deal with the former two problems but common hardware platforms cannot support them efficiently. To tackle all these challenges and make the training phase of CNNs benefit from the low-bit format, we propose a low-bit training framework for CNNs to pursue a better tradeoff between accuracy and energy efficiency: 1) we adopt element-wise scaling to increase the dynamic range of data representation, which significantly reduces the quantization error; 2) group-wise scaling with hardware friendly factor format is designed to reduce the element-wise exponent bits without degrading the accuracy; and 3) we design the customized hardware unit that implements the low-bit tensor convolution arithmetic with our multilevel scaling data format. Experiments show that our framework achieves a superior tradeoff between the accuracy and the bit-width than previous low-bit training studies. For training various models on CIFAR-10, using 1-bit mantissa and 2-bit exponent is adequate to keep the accuracy loss within 1%. On larger datasets like ImageNet, using 4-bit mantissa and 2-bit exponent is adequate. Through the energy consumption simulation of the whole network, we can see that training a variety of models with our framework could achieve <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$4.9\times $ </tex-math></inline-formula> – <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$10.2\times $ </tex-math></inline-formula> higher energy efficiency than full-precision arithmetic.

Parameter Selection of Direct Modulation Semiconductor Laser for Shaping Current Based on Convolutional Neural Network

The shaping current technology can efficiently and low-costly suppress the relaxation oscillations (ROs) of the direct modulation semiconductor laser (DML) for the high-performance optic system. The parameter selection is the key problem to precisely constructing the injection current form to obtain the desired output waveform. A novel framework based on convolutional neural network (CNN) is proposed to predict the shaped current parameters avoiding the time-consuming and computationally complex problems with analytical solutions. In the network training, batch and min-max normalizations are adopted to optimize neural networks, which aim to accelerate the convergence and improve their approximation ability. The trained inverse CNN named by feeding into the desired data samples from DML output waveform is used to achieve parameter selection for constructing the injection current. Also, the trained forward CNN would verify the validity of selected parameters responding to the output waveform, and get the unique corresponding relationship between them. Simulation results own high agreement with the theoretical values and show that the CNN models provide a powerful tool to select parameters of shaped current with accurate and fast capabilities.

Research of Obfuscated Malware with a Capsule Neural Network

The paper presents the results of a research of using transfer training of the capsule neural network to detect malware. The research was carried out on the basis of the source code of malware using the context-triggered piecewise hashing method. The source codes of malware were obtained from public sources of software. Verification of the capsule neural network learning results was carried out using a trained convolutional neural network, and publicly available sources of test to malware. The research was conducted on six types of malware. Software source code, part of capsule neural network training datasets, pre-trained capsule neural network, and full research are publicly available at https://github.com/T-JN.

A Fast, Accurate, and Comprehensive PPA Estimation of Convolutional Hardware Accelerators

Convolutional Neural Networks (CNN) are widely adopted for Machine Learning (ML) tasks, such as classification and computer vision. GPUs became the reference platforms for both training and inference phases of CNNs due to their tailored architecture to the CNN operators. However, GPUs are power-hungry architectures. A path to enable the deployment of CNNs in energy-constrained devices is adopting hardware accelerators for the inference phase. However, the literature presents gaps regarding analyses and comparisons of these accelerators to evaluate Power-Performance-Area (PPA) trade-offs. Typically, the literature estimates PPA from the number of executed operations during the inference phase, such as the number of MACs, which may not be a good proxy for PPA. Thus, it is necessary to deliver accurate hardware estimations, enabling design space exploration (DSE) to deploy CNNs according to the design constraints. This work proposes a fast and accurate DSE approach for CNNs using an analytical model fitted from the physical synthesis of hardware accelerators. The model is integrated with CNN frameworks, like TensorFlow, to generate accurate results. The analytic model estimates area, performance, power, energy, and memory accesses. The observed average error comparing the analytical model to the data obtained from the physical synthesis is smaller than 7%.

Image based anthracnose and red-rust leaf disease detection using deep learning

Deep residual learning frameworks have achieved great success in image classification. This article presents the use of transfer learning which is applied on mango leaf image dataset for its disease’s detection. New methodology and training have been used to facilitate the easy and rapid implementation of the mango leaf disease detection system in practice. Proposed system can be used to identify the mango leaf for whether it is healthy or infected with the diseases like anthracnose or red rust. This paper describes all the steps which are considered during the experimentation and design. These steps include leaf image data collection, its preparation, data assessment by agricultural experts, and selection and tranning of deep neural network architectures. A deep residual framework, residual neural network (ResNET), was used to perform deep convolutional neural network training. ResNETs are easy to optimize and can achieve better accuracies. The experimental results obtained from “ResNET architectures, such as ResNet18, ResNet34, ResNet50, and ResNet101” show the accuracies from 94% to 98%. ResNET18 architecture selected from above for system design as it gives 98% accuracy for mango leaf disease’s detection. System will help farmers to identify leaf diseases in quick and efficient manner and facilitate decision-making in this front.

Cell2Grid: an efficient, spatial, and convolutional neural network-ready representation of cell segmentation data.

Cell segmentation algorithms are commonly used to analyze large histologic images as they facilitate interpretation, but on the other hand they complicate hypothesis-free spatial analysis. Therefore, many applications train convolutional neural networks (CNNs) on high-resolution images that resolve individual cells instead, but their practical application is severely limited by computational resources. In this work, we propose and investigate an alternative spatial data representation based on cell segmentation data for direct training of CNNs. We introduce and analyze the properties of Cell2Grid, an algorithm that generates compact images from cell segmentation data by placing individual cells into a low-resolution grid and resolves possible cell conflicts. For evaluation, we present a case study on colorectal cancer relapse prediction using fluorescent multiplex immunohistochemistry images. We could generate Cell2Grid images at resolution that were 100 times smaller than the original ones. Cell features, such as phenotype counts and nearest-neighbor cell distances, remain similar to those of original cell segmentation tables ( ). These images could be directly fed to a CNN for predicting colon cancer relapse. Our experiments showed that test set error rate was reduced by 25% compared with CNNs trained on images rescaled to with bilinear interpolation. Compared with images at resolution (bilinear rescaling), our method reduced CNN training time by 85%. Cell2Grid is an efficient spatial data representation algorithm that enables the use of conventional CNNs on cell segmentation data. Its cell-based representation additionally opens a door for simplified model interpretation and synthetic image generation.

Extracting 3-D Radar Features to Improve Quantitative Precipitation Estimation in Complex Terrain based on Deep Learning Neural Networks

Abstract This paper proposes a new quantitative precipitation estimation (QPE) technique to provide accurate rainfall estimates in complex terrain, where conventional QPE has limitations. The operational radar QPE in Taiwan is mainly based on the simplified relationship between radar reflectivity Z and rain rate R [R(Z) relation] and only utilizes the single-point lowest available echo to estimate rain rates, leading to low accuracy in complex terrain. Here, we conduct QPE using deep learning that extracts features from 3-D radar reflectivities to address the above issues. Convolutional neural networks (CNN) are used to analyze contoured frequency by altitude diagrams (CFADs) to generate the QPE. CNN models are trained on existing rain gauges in northern and eastern Taiwan with the three-year data during 2015–17 and validated and tested using 2018 data. The weights of heavy rains (≧10 mm h-1) are increased in the model loss calculation to handle the unbalanced rainfall data and improve accuracy. Results show that the CNN outperforms the R(Z) relation based on the 2018 rain-gauge data. Furthermore, this research proposes methods to conduct 2-D gridded QPE at every pixel by blending estimates from various trained CNN models. Verification based on independent rain gauges shows that the CNN QPE solves the underestimation of the R(Z) relation in mountainous areas. Case studies are presented to visualize the results, showing that the CNN QPE generates better small-scale rainfall features and more accurate precipitation information. This deep learning QPE technique may be helpful for the disaster prevention of small-scale flash floods in complex terrain.

Training CNN Classifiers Solely on Webly Data

Abstract Real life applications of deep learning (DL) are often limited by the lack of expert labeled data required to effectively train DL models. Creation of such data usually requires substantial amount of time for manual categorization, which is costly and is considered to be one of the major impediments in development of DL methods in many areas. This work proposes a classification approach which completely removes the need for costly expert labeled data and utilizes noisy web data created by the users who are not subject matter experts. The experiments are performed with two well-known Convolutional Neural Network (CNN) architectures: VGG16 and ResNet50 trained on three randomly collected Instagram-based sets of images from three distinct domains: metropolitan cities, popular food and common objects - the last two sets were compiled by the authors and made freely available to the research community. The dataset containing common objects is a webly counterpart of PascalVOC2007 set. It is demonstrated that despite significant amount of label noise in the training data, application of proposed approach paired with standard training CNN protocol leads to high classification accuracy on representative data in all three above-mentioned domains. Additionally, two straightforward procedures of automatic cleaning of the data, before its use in the training process, are proposed. Apparently, data cleaning does not lead to improvement of results which suggests that the presence of noise in webly data is actually helpful in learning meaningful and robust class representations. Manual inspection of a subset of web-based test data shows that labels assigned to many images are ambiguous even for humans. It is our conclusion that for the datasets and CNN architectures used in this paper, in case of training with webly data, a major factor contributing to the final classification accuracy is representativeness of test data rather than application of data cleaning procedures.