Train Deep Convolutional Neural Networks Research Articles

UV‐flashover evaluation of porcelain insulators based on deep learning

Based on the analysis of the principle and structure of the convolutional neural network (CNN) model in a deep learning theory system, an intelligent method for judging the flashover of a porcelain insulator with ultraviolet discharge is proposed. In this method, the porcelain insulator chip was subjected to power frequency flashover testing, and the ultraviolet spectra of different discharge states without discharge, weak coronal discharge, and strong spark discharge were captured by FILIN UV imager. The Alexnet deep convolution neural network model was used to predict the discharge state of the UV spectrum for classification training and identification assessment. The new method doesn't use UV imaging to detect flashover warnings. It is necessary to extract the characteristics of the UV spectra and leakage current parameters manually. The multi-layer combination of UV imaging method with end-to-end autonomous learning with deep learning training and the adopted test method provided classification identification, through a large number of UV images in the deep CNN training. This allowed flashover evaluation of abstract feature parameters of the independent extraction. The results show that this method has the advantages of high accuracy, and provides a new idea for the intelligent detection of UV flashover.

Unsupervised pixel-wise classification for Chaetoceros image segmentation

Abstract Chaetoceros is a dominant genus of marine planktonic diatoms with worldwide distribution. Due to the difficulty of extracting setae from Chaetoceros images, automatic segmentation of Chaetoceros is still a challenging task. In this paper, we address this difficult task by regarding the whole segmentation process as unsupervised pixel-wise classification without human participation. First, we automatically produce positive (object) and negative (background) samples for follow-up training, by combining the advantages of two image processing algorithms: Grayscale Surface Direction Angle Model (GSDAM) for extracting setae information and Canny for detecting cell edges from low-contrast and strong-noisy microscopic images. Second, we develop pixel-wise training by using the produced samples in the training process of Deep Convolutional Neural Network (DCNN). At last, the trained DCNN is used to label other pixels into object and background for final segmentation. We compare our method with eight mainstream segmentation approaches: Otsu’s thresholding, Canny, Watershed, Mean Shift, gPb-owt-ucm, Normalized Cut, Efficient Graph-based method and GSDAM. To objectively evaluate segmentation results, we apply six well-known evaluation indexes. Experimental results on a new Chaetoceros image dataset with human labelled ground truth show that our method outperforms the eight mainstream segmentation methods in terms of both quantitative and qualitative evaluation.

Training Deep Spiking Convolutional Neural Networks With STDP-Based Unsupervised Pre-training Followed by Supervised Fine-Tuning.

Spiking Neural Networks (SNNs) are fast becoming a promising candidate for brain-inspired neuromorphic computing because of their inherent power efficiency and impressive inference accuracy across several cognitive tasks such as image classification and speech recognition. The recent efforts in SNNs have been focused on implementing deeper networks with multiple hidden layers to incorporate exponentially more difficult functional representations. In this paper, we propose a pre-training scheme using biologically plausible unsupervised learning, namely Spike-Timing-Dependent-Plasticity (STDP), in order to better initialize the parameters in multi-layer systems prior to supervised optimization. The multi-layer SNN is comprised of alternating convolutional and pooling layers followed by fully-connected layers, which are populated with leaky integrate-and-fire spiking neurons. We train the deep SNNs in two phases wherein, first, convolutional kernels are pre-trained in a layer-wise manner with unsupervised learning followed by fine-tuning the synaptic weights with spike-based supervised gradient descent backpropagation. Our experiments on digit recognition demonstrate that the STDP-based pre-training with gradient-based optimization provides improved robustness, faster (~2.5 ×) training time and better generalization compared with purely gradient-based training without pre-training.

The FIRST Classifier: compact and extended radio galaxy classification using deep Convolutional Neural Networks

Upcoming surveys with new radio observatories such as the Square Kilometer Array will generate a wealth of imaging data containing large numbers of radio galaxies. Different classes of radio galaxies can be used as tracers of the cosmic environment, including the dark matter density field, to address key cosmological questions. Classifying these galaxies based on morphology is thus an important step toward achieving the science goals of next generation radio surveys. Radio galaxies have been traditionally been classified as Fanaroff-Riley (FR) I and II, although some exhibit more complex 'bent' morphologies arising from environmental factors or intrinsic properties. In this work we present the FIRST Classifier, an on-line system for automated classification of Compact and Extended radio sources. We developed the FIRST Classifier based on a trained Deep Convolutional Neural Network Model to automate the morphological classification of compact and extended radio sources observed in the FIRST radio survey. Our model achieved an overall accuracy of 97% and a recall of 98%, 100%, 98% and 93% for Compact, BENT, FRI and FRII galaxies respectively. The current version of the FIRST classifier is able to predict the morphological class for a single source or for a list of sources as Compact or Extended (FRI, FRII and BENT).

Landscape Classification with Deep Neural Networks

The application of deep learning, specifically deep convolutional neural networks (DCNNs), to the classification of remotely-sensed imagery of natural landscapes has the potential to greatly assist in the analysis and interpretation of geomorphic processes. However, the general usefulness of deep learning applied to conventional photographic imagery at a landscape scale is, at yet, largely unproven. If DCNN-based image classification is to gain wider application and acceptance within the geoscience community, demonstrable successes need to be coupled with accessible tools to retrain deep neural networks to discriminate landforms and land uses in landscape imagery. Here, we present an efficient approach to train/apply DCNNs with/on sets of photographic images, using a powerful graphical method called a conditional random field (CRF), to generate DCNN training and testing data using minimal manual supervision. We apply the method to several sets of images of natural landscapes, acquired from satellites, aircraft, unmanned aerial vehicles, and fixed camera installations. We synthesize our findings to examine the general effectiveness of transfer learning to landscape-scale image classification. Finally, we show how DCNN predictions on small regions of images might be used in conjunction with a CRF for highly accurate pixel-level classification of images.

Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning

Having accurate, detailed, and up-to-date information about the location and behavior of animals in the wild would improve our ability to study and conserve ecosystems. We investigate the ability to automatically, accurately, and inexpensively collect such data, which could help catalyze the transformation of many fields of ecology, wildlife biology, zoology, conservation biology, and animal behavior into "big data" sciences. Motion-sensor "camera traps" enable collecting wildlife pictures inexpensively, unobtrusively, and frequently. However, extracting information from these pictures remains an expensive, time-consuming, manual task. We demonstrate that such information can be automatically extracted by deep learning, a cutting-edge type of artificial intelligence. We train deep convolutional neural networks to identify, count, and describe the behaviors of 48 species in the 3.2 million-image Snapshot Serengeti dataset. Our deep neural networks automatically identify animals with >93.8% accuracy, and we expect that number to improve rapidly in years to come. More importantly, if our system classifies only images it is confident about, our system can automate animal identification for 99.3% of the data while still performing at the same 96.6% accuracy as that of crowdsourced teams of human volunteers, saving >8.4 y (i.e., >17,000 h at 40 h/wk) of human labeling effort on this 3.2 million-image dataset. Those efficiency gains highlight the importance of using deep neural networks to automate data extraction from camera-trap images, reducing a roadblock for this widely used technology. Our results suggest that deep learning could enable the inexpensive, unobtrusive, high-volume, and even real-time collection of a wealth of information about vast numbers of animals in the wild.

Predicting RNA-protein binding sites and motifs through combining local and global deep convolutional neural networks.

RNA-binding proteins (RBPs) take over 5-10% of the eukaryotic proteome and play key roles in many biological processes, e.g. gene regulation. Experimental detection of RBP binding sites is still time-intensive and high-costly. Instead, computational prediction of the RBP binding sites using patterns learned from existing annotation knowledge is a fast approach. From the biological point of view, the local structure context derived from local sequences will be recognized by specific RBPs. However, in computational modeling using deep learning, to our best knowledge, only global representations of entire RNA sequences are employed. So far, the local sequence information is ignored in the deep model construction process. In this study, we present a computational method iDeepE to predict RNA-protein binding sites from RNA sequences by combining global and local convolutional neural networks (CNNs). For the global CNN, we pad the RNA sequences into the same length. For the local CNN, we split a RNA sequence into multiple overlapping fixed-length subsequences, where each subsequence is a signal channel of the whole sequence. Next, we train deep CNNs for multiple subsequences and the padded sequences to learn high-level features, respectively. Finally, the outputs from local and global CNNs are combined to improve the prediction. iDeepE demonstrates a better performance over state-of-the-art methods on two large-scale datasets derived from CLIP-seq. We also find that the local CNN runs 1.8 times faster than the global CNN with comparable performance when using GPUs. Our results show that iDeepE has captured experimentally verified binding motifs. https://github.com/xypan1232/iDeepE. Supplementary data are available at Bioinformatics online.

Deep convolutional neural network training enrichment using multi-view object-based analysis of Unmanned Aerial systems imagery for wetlands classification

Deep convolutional neural network (DCNN) requires massive training datasets to trigger its image classification power, while collecting training samples for remote sensing application is usually an expensive process. When DCNN is simply implemented with traditional object-based image analysis (OBIA) for classification of Unmanned Aerial systems (UAS) orthoimage, its power may be undermined if the number training samples is relatively small. This research aims to develop a novel OBIA classification approach that can take advantage of DCNN by enriching the training dataset automatically using multi-view data. Specifically, this study introduces a Multi-View Object-based classification using Deep convolutional neural network (MODe) method to process UAS images for land cover classification. MODe conducts the classification on multi-view UAS images instead of directly on the orthoimage, and gets the final results via a voting procedure. 10-fold cross validation results show the mean overall classification accuracy increasing substantially from 65.32%, when DCNN was applied on the orthoimage to 82.08% achieved when MODe was implemented. This study also compared the performances of the support vector machine (SVM) and random forest (RF) classifiers with DCNN under traditional OBIA and the proposed multi-view OBIA frameworks. The results indicate that the advantage of DCNN over traditional classifiers in terms of accuracy is more obvious when these classifiers were applied with the proposed multi-view OBIA framework than when these classifiers were applied within the traditional OBIA framework.

Image Classification Based on the Boost Convolutional Neural Network

Convolutional neural networks (CNNs), which are composed of multiple processing layers to learn the representations of data with multiple abstract levels, are the most successful machine learning models in recent years. However, these models can have millions of parameters and many layers, which are difficult to train, and sometimes several days or weeks are required to tune the parameters. Within this paper, we present the usage of a trained deep convolutional neural network model to extract the features of the images, and then, used the AdaBoost algorithm to assemble the Softmax classifiers into recognizable images. This method resulted in a 3% increase of accuracy of the trained CNN models, and dramatically reduced the retraining time cost, and thus, it has good application prospects.

Deep Multi-Instance Multi-Label Learning for Image Annotation

Multi-Instance Multi-Label learning (MIML) is a popular framework for supervised classification where an example is described by multiple instances and associated with multiple labels. Previous MIML approaches have focused on predicting labels for instances. The idea of tackling the problem is to identify its equivalence in the traditional supervised learning framework. Motivated by the recent advancement in deep learning, in this paper, we still consider the problem of predicting labels and attempt to model deep learning in MIML learning framework. The proposed approach enables us to train deep convolutional neural network with images from social networks where images are well labeled, even labeled with several labels or uncorrelated labels. Experiments on real-world datasets demonstrate the effectiveness of our proposed approach.

Multi-task transfer learning deep convolutional neural network: application to computer-aided diagnosis of breast cancer on mammograms

Transfer learning in deep convolutional neural networks (DCNNs) is an important step in its application to medical imaging tasks. We propose a multi-task transfer learning DCNN with the aim of translating the ‘knowledge’ learned from non-medical images to medical diagnostic tasks through supervised training and increasing the generalization capabilities of DCNNs by simultaneously learning auxiliary tasks. We studied this approach in an important application: classification of malignant and benign breast masses. With Institutional Review Board (IRB) approval, digitized screen-film mammograms (SFMs) and digital mammograms (DMs) were collected from our patient files and additional SFMs were obtained from the Digital Database for Screening Mammography. The data set consisted of 2242 views with 2454 masses (1057 malignant, 1397 benign). In single-task transfer learning, the DCNN was trained and tested on SFMs. In multi-task transfer learning, SFMs and DMs were used to train the DCNN, which was then tested on SFMs. N-fold cross-validation with the training set was used for training and parameter optimization. On the independent test set, the multi-task transfer learning DCNN was found to have significantly (p = 0.007) higher performance compared to the single-task transfer learning DCNN. This study demonstrates that multi-task transfer learning may be an effective approach for training DCNN in medical imaging applications when training samples from a single modality are limited.

A deep convolutional neural network for classification of red blood cells in sickle cell anemia.

Sickle cell disease (SCD) is a hematological disorder leading to blood vessel occlusion accompanied by painful episodes and even death. Red blood cells (RBCs) of SCD patients have diverse shapes that reveal important biomechanical and bio-rheological characteristics, e.g. their density, fragility, adhesive properties, etc. Hence, having an objective and effective way of RBC shape quantification and classification will lead to better insights and eventual better prognosis of the disease. To this end, we have developed an automated, high-throughput, ex-vivo RBC shape classification framework that consists of three stages. First, we present an automatic hierarchical RBC extraction method to detect the RBC region (ROI) from the background, and then separate touching RBCs in the ROI images by applying an improved random walk method based on automatic seed generation. Second, we apply a mask-based RBC patch-size normalization method to normalize the variant size of segmented single RBC patches into uniform size. Third, we employ deep convolutional neural networks (CNNs) to realize RBC classification; the alternating convolution and pooling operations can deal with non-linear and complex patterns. Furthermore, we investigate the specific shape factor quantification for the classified RBC image data in order to develop a general multiscale shape analysis. We perform several experiments on raw microscopy image datasets from 8 SCD patients (over 7,000 single RBC images) through a 5-fold cross validation method both for oxygenated and deoxygenated RBCs. We demonstrate that the proposed framework can successfully classify sickle shape RBCs in an automated manner with high accuracy, and we also provide the corresponding shape factor analysis, which can be used synergistically with the CNN analysis for more robust predictions. Moreover, the trained deep CNN exhibits good performance even for a deoxygenated dataset and distinguishes the subtle differences in texture alteration inside the oxygenated and deoxygenated RBCs.

Training Deep Convolutional Neural Networks with Resistive Cross-Point Devices

In a previous work we have detailed the requirements for obtaining maximal deep learning performance benefit by implementing fully connected deep neural networks (DNN) in the form of arrays of resistive devices. Here we extend the concept of Resistive Processing Unit (RPU) devices to convolutional neural networks (CNNs). We show how to map the convolutional layers to fully connected RPU arrays such that the parallelism of the hardware can be fully utilized in all three cycles of the backpropagation algorithm. We find that the noise and bound limitations imposed by the analog nature of the computations performed on the arrays significantly affect the training accuracy of the CNNs. Noise and bound management techniques are presented that mitigate these problems without introducing any additional complexity in the analog circuits and that can be addressed by the digital circuits. In addition, we discuss digitally programmable update management and device variability reduction techniques that can be used selectively for some of the layers in a CNN. We show that a combination of all those techniques enables a successful application of the RPU concept for training CNNs. The techniques discussed here are more general and can be applied beyond CNN architectures and therefore enables applicability of the RPU approach to a large class of neural network architectures.

Fusion of Deep Convolutional Neural Networks for Land Cover Classification of High-Resolution Imagery

Deep convolutional neural networks (DCNNs) have recently emerged as a dominant paradigm for machine learning in a variety of domains. However, acquiring a suitably large data set for training DCNN is often a significant challenge. This is a major issue in the remote sensing domain, where we have extremely large collections of satellite and aerial imagery, but lack the rich label information that is often readily available for other image modalities. In this letter, we investigate the use of DCNN for land–cover classification in high-resolution remote sensing imagery. To overcome the lack of massive labeled remote-sensing image data sets, we employ two techniques in conjunction with DCNN: transfer learning (TL) with fine-tuning and data augmentation tailored specifically for remote sensing imagery. TL allows one to bootstrap a DCNN while preserving the deep visual feature extraction learned over an image corpus from a different image domain. Data augmentation exploits various aspects of remote sensing imagery to dramatically expand small training image data sets and improve DCNN robustness for remote sensing image data. Here, we apply these techniques to the well-known UC Merced data set to achieve the land–cover classification accuracies of 97.8 ± 2.3%, 97.6 ± 2.6%, and 98.5 ± 1.4% with CaffeNet, GoogLeNet, and ResNet, respectively.

Video Object Segmentation Through Deep Convolutional Networks

Video object segmentation of real-world scenes is a challenging task due to the dynamic environment change such as uneven illumination, object articulation, as well as camera motion. In this paper, we proposed a method to combine semantic regions by trained deep convolutional neural networks with saliency map and motion cues to segment foreground objects in video sequences. Firstly, the parameters of the deep convolutional networks are trained using PASCAL VOC dataset with human annotations. The training process consists of forward inference and background learning stages. The learning process employs the standard stochastic gradient descending algorithm. The number of epochs during training process is fixed at 7. Secondly, the trained convolutional networks are used to predict the semantic labels of a real-world video sequence at per-frame level. The inferenced semantic region is combined with the saliency map through Markov random field to derive foreground objects at each frame of the video. To access the segmentation performance of the proposed algorithm, we test the proposed algorithm using a video sequence from FBMS motion segmentation benchmarks and compare the segmentation accuracy with state-of-the-art video object segmentation algorithms.

Mixtures of Lightweight Deep Convolutional Neural Networks: Applied to Agricultural Robotics

We propose a novel approach for training deep convolutional neural networks (DCNNs) that allows us to tradeoff complexity and accuracy to learn lightweight models suitable for robotic platforms such as AgBot II (which performs automated weed management). Our approach consists of three stages, the first is to adapt a pre-trained model to the task at hand. This provides state-of-the-art performance but at the cost of high computational complexity resulting in a low frame rate of just 0.12 frames per second (fps). Second, we use the adapted model and employ model compression techniques to learn a lightweight DCNN that is less accurate but has two orders of magnitude fewer parameters. Third, $K$ lightweight models are combined as a mixture model to further enhance the performance of the lightweight models. Applied to the challenging task of weed segmentation, we improve the accuracy from 85.9%, using a traditional approach, to 93.9% by adapting a complicated pre-trained DCNN with 25M parameters (Inception-v3). The downside to this adapted model, Adapted-IV3, is that it can only process 0.12 fps. To make this approach fast while still retaining accuracy, we learn lightweight DCNNs which when combined can achieve accuracy greater than 90% while using considerably fewer parameters capable of processing between 1.07 and 1.83 fps, up to an order of magnitude faster and up to an order of magnitude fewer parameters.

Handcrafted vs. non-handcrafted features for computer vision classification

This work presents a generic computer vision system designed for exploiting trained deep Convolutional Neural Networks (CNN) as a generic feature extractor and mixing these features with more traditional hand-crafted features. Such a system is a single structure that can be used for synthesizing a large number of different image classification tasks. Three substructures are proposed for creating the generic computer vision system starting from handcrafted and non-handcrafter features: i)one that remaps the output layer of a trained CNN to classify a different problem using an SVM; ii) a second for exploiting the output of the penultimate layer of a trained CNN as a feature vector to feed an SVM; and iii) a third for merging the output of some deep layers, applying a dimensionality reduction method, and using these features as the input to an SVM. The application of feature transform techniques to reduce the dimensionality of feature sets coming from the deep layers represents one of the main contributions of this paper. Three approaches are used for the non-handcrafted features: deep transfer learning features based on convolutional neural networks (CNN), principal component analysis network (PCAN), and the compact binary descriptor (CBD). For the handcrafted features, a wide variety of state-of-the-art algorithms are considered: Local Ternary Patterns, Local Phase Quantization, Rotation Invariant Co-occurrence Local Binary Patterns, Completed Local Binary Patterns, Rotated local binary pattern image, Globally Rotation Invariant Multi-scale Co-occurrence Local Binary Pattern, and several others. The computer vision system based on the proposed approach was tested on many different datasets, demonstrating the generalizability of the proposed approach thanks to the strong performance recorded. The Wilcoxon signed rank test is used to compare the different methods; moreover, the independence of the different methods is studied using the Q-statistic. To facilitate replication of our experiments, the MATLAB source code will be available at (https://www.dropbox.com/s/bguw035yrqz0pwp/ElencoCode.docx?dl=0).

Learning Dual Multi-Scale Manifold Ranking for Semantic Segmentation of High-Resolution Images

Semantic image segmentation has recently witnessed considerable progress by training deep convolutional neural networks (CNNs). The core issue of this technique is the limited capacity of CNNs to depict visual objects. Existing approaches tend to utilize approximate inference in a discrete domain or additional aides and do not have a global optimum guarantee. We propose the use of the multi-label manifold ranking (MR) method in solving the linear objective energy function in a continuous domain to delineate visual objects and solve these problems. We present a novel embedded single stream optimization method based on the MR model to avoid approximations without sacrificing expressive power. In addition, we propose a novel network, which we refer to as dual multi-scale manifold ranking (DMSMR) network, that combines the dilated, multi-scale strategies with the single stream MR optimization method in the deep learning architecture to further improve the performance. Experiments on high resolution images, including close-range and remote sensing datasets, demonstrate that the proposed approach can achieve competitive accuracy without additional aides in an end-to-end manner.

Training Deep Convolutional Neural Networks for Land–Cover Classification of High-Resolution Imagery

Deep convolutional neural networks (DCNNs) have recently emerged as a dominant paradigm for machine learning in a variety of domains. However, acquiring a suitably large data set for training DCNN is often a significant challenge. This is a major issue in the remote sensing domain, where we have extremely large collections of satellite and aerial imagery, but lack the rich label information that is often readily available for other image modalities. In this letter, we investigate the use of DCNN for land-cover classification in high-resolution remote sensing imagery. To overcome the lack of massive labeled remote-sensing image data sets, we employ two techniques in conjunction with DCNN: transfer learning (TL) with fine-tuning and data augmentation tailored specifically for remote sensing imagery. TL allows one to bootstrap a DCNN while preserving the deep visual feature extraction learned over an image corpus from a different image domain. Data augmentation exploits various aspects of remote sensing imagery to dramatically expand small training image data sets and improve DCNN robustness for remote sensing image data. Here, we apply these techniques to the well-known UC Merced data set to achieve the land-cover classification accuracies of 97.8 ± 2.3%, 97.6 ± 2.6%, and 98.5 ± 1.4% with CaffeNet, GoogLeNet, and ResNet, respectively.

Training DCNN by Combining Max-Margin, Max-Correlation Objectives, and Correntropy Loss for Multilabel Image Classification.

In this paper, we build a multilabel image classifier using a general deep convolutional neural network (DCNN). We propose a novel objective function that consists of three parts, i.e., max-margin objective, max-correlation objective, and correntropy loss. The max-margin objective explicitly enforces that the minimum score of positive labels must be larger than the maximum score of negative labels by a predefined margin, which not only improves accuracies of the multilabel classifier, but also eases the threshold determination. The max-correlation objective can make the DCNN model learn a latent semantic space, which maximizes the correlations between the feature vectors of the training samples and their corresponding ground-truth label vectors projected into this space. Instead of using the traditional softmax loss, we adopt the correntropy loss from the information theory field to minimize the training errors of the DCNN model. The proposed framework can be end-to-end trained. Comprehensive experimental evaluations on Pascal VOC 2007 and MIR Flickr 25K multilabel benchmark data sets with four DCNN models, i.e., AlexNet, VGG-16, GoogLeNet, and ResNet demonstrate that the proposed objective function can remarkably improve the performance accuracies of a DCNN model for the task of multilabel image classification.