Search

The objective of this study was to develop a deep learning-based tree species identification model using pollen grain images taken with a camera mounted on an optical microscope. From five focal points, we took photographs of pollen collected from tree species widely distributed in the Japanese archipelago, and we used these to produce pollen images. We used Caffe as the deep learning framework and AlexNet and GoogLeNet as the deep learning algorithms. We constructed four learning models that combined two learning patterns, one for focal point images with data augmentation, for which the training and test data were the same, and the other without data augmentation, for which they were not the same. The performance of the proposed model was evaluated according to the MCC and F score. The most accurate classification model was based on the GoogLeNet algorithm, with data augmentation after 200 epochs. Tree species identification accuracy varied depending on the focal point, even for the same pollen grain, and images focusing on the pollen surface tended to be more accurately classified than those focusing on the pollen outline and membrane structure. Castanea crenata, Fraxinus sieboldiana, and Quercus crispula pollen grains were classified with the highest accuracy, whereas Gamblea innovans, Carpinus tschonoskii, Cornus controversa, Fagus japonica, Quercus serrata, and Quercus sessilifolia showed the lowest classification accuracy. Future studies should consider application to fossil pollen in sediments and state-of-the-art deep learning algorithms. [ABSTRACT FROM AUTHOR]

Convolutional Neural Networks (CNNs) have become the default paradigm for addressing classification problems, especially, but not only, in image recognition. This is mainly due to their high success rate. Although a number of approaches currently apply deep learning to the 3D shape recognition problem, they are either too slow for online use or too error-prone. To fill this gap, we propose 3DSliceLeNet, a deep learning architecture for point cloud classification. Our proposal converts the input point clouds into a two-dimensional representation by performing a slicing process and projecting the points to the principal planes, thus generating images that are used by the convolutional architecture. 3DSliceLeNet successfully achieves both high accuracy and low computational cost. A dense set of experiments has been conducted to validate our system under the ModelNet challenge, a large-scale 3D Computer Aided Design (CAD) model dataset. Our proposal achieves a success rate of 94.37% and an Area under Curve (AUC) of 0.978 on the ModelNet-10 classification task.

The objective of this study was to verify the accuracy of tree species identification using deep learning with leaf images of broadleaf and coniferous trees in outdoor photographs. For each of 12 broadleaf and eight coniferous tree species, we acquired 300 photographs of leaves and used those to produce 72,000 256 × 256-pixel images. We used Caffe as the deep learning framework and AlexNet and GoogLeNet as the deep learning algorithms. We constructed four learning models that combined two learning patterns: one for individual classification of 20 species and the other for two-group classification (broadleaf vs. coniferous trees), with and without data augmentation, respectively. The performance of the proposed model was evaluated according to the MCC and F-score. Both classification models exhibited very high accuracy for all learning patterns; the highest MCC was 0.997 for GoogLeNet with data augmentation. The classification accuracy was higher for broadleaf trees when the model was trained using broadleaf only; for coniferous trees, the classification accuracy was higher when the model was trained using both tree types simultaneously than when it was trained using coniferous trees only. [ABSTRACT FROM AUTHOR]

The objective of this study was to develop a deep learning-based tree species identification model using pollen grain images taken with a camera mounted on an optical microscope. From five focal points, we took photographs of pollen collected from tree species widely distributed in the Japanese archipelago, and we used these to produce pollen images. We used Caffe as the deep learning framework and AlexNet and GoogLeNet as the deep learning algorithms. We constructed four learning models that combined two learning patterns, one for focal point images with data augmentation, for which the training and test data were the same, and the other without data augmentation, for which they were not the same. The performance of the proposed model was evaluated according to the MCC and F score. The most accurate classification model was based on the GoogLeNet algorithm, with data augmentation after 200 epochs. Tree species identification accuracy varied depending on the focal point, even for the same pollen grain, and images focusing on the pollen surface tended to be more accurately classified than those focusing on the pollen outline and membrane structure. Castanea crenata, Fraxinus sieboldiana, and Quercus crispula pollen grains were classified with the highest accuracy, whereas Gamblea innovans, Carpinus tschonoskii, Cornus controversa, Fagus japonica, Quercus serrata, and Quercus sessilifolia showed the lowest classification accuracy. Future studies should consider application to fossil pollen in sediments and state-of-the-art deep learning algorithms.

Based on the growing demand for neural network technologies, various neural network inference engines are being developed. However, each inference engine has its own neural network storage format. There is a growing demand for standardization to solve this problem. This study presents interworking techniques for ensuring the compatibility of neural networks and data among the various deep learning frameworks. The proposed technique standardizes the graphic expression grammar and learning data storage format using the Neural Network Exchange Format (NNEF) of Khronos. The proposed converter includes a lexical, syntax, and parser. This NNEF parser converts neural network information into a parsing tree and quantizes data. To validate the proposed system, we verified that MNIST is immediately executed by importing AlexNet's neural network and learned data. Therefore, this study contributes an efficient design technique for a converter that can execute a neural network and learned data in various frameworks regardless of the storage format of each framework.

This article investigates the history of coffee culture across three continents during the Fascist ventennio (1922–45.) By using the novel framework of coffee, from the bean in the field to the machine in the caffè, it connects interwar histories that previously have been explored independently. Specifically, it examines the transnational economics of coffee bean trade routes and the colonial imagery of coffee advertising to argue that caffès emerged as key sites for promoting the Fascist imperial projects in East Africa – an architectural and artistic legacy that remains in place today. Ultimately, this trajectory broadens the way that we understand how food and farming became politicised during the Fascist period. By untangling the interwar trade of beans and bodies between Italy, Brazil, and Ethiopia, this article brings to light an untold story of caffeinated imperial aggression and resistance. [ABSTRACT FROM AUTHOR]

With the success of deep learning (DL) methods in diverse application domains, several deep learning software frameworks have been proposed to facilitate the usage of these methods. By knowing the frameworks which are employed in big data analysis, the analysis process will be more efficient in terms of time and accuracy. Thus, benchmarking DL software frameworks is in high demand. This paper presents a comparative study of deep learning frameworks, namely Caffe and TensorFlow on performance metrics: runtime performance and accuracy. This study is performed with several datasets, such as LeNet MNIST classification model, CIFAR-10 image recognition datasets and message passing interface (MPI) parallel matrix-vector multiplication. We evaluate the performance of the above frameworks when employed on machines of Intel Xeon Phi 7210. In this study, the use of vectorization, OpenMP parallel processing, and MPI are examined to improve the performance of deep learning frameworks. The experimental results show the accuracy comparison between the number of iterations of the test in the training model and the training time on the different machines before and after optimization. In addition, an experiment on two multi-nodes of Xeon Phi is performed. The experimental results also show the optimization of Xeon Phi is beneficial to the Caffe and TensorFlow frameworks. [ABSTRACT FROM AUTHOR]

In the eighteenth century the scientists had very different opinions about the new foods that had spread in Europe after the discovery of America. Also in Verona there was a lot of interest in this matter and an illustrious doctor (Giovanni Dalla Bona) published in 1751 an interesting book on this subject (in particular about coffee and chocolate). He took an intermediate position between those who considered these new foods and drinks harmful to health and those who considered them healthy. Dalla Bona, basing on the science of his time, came to a conclusion that coincides with the common sense: it is necessary to distinguish case by case (coffee, for example, is good for some but bad for others) and, above all, to avoid the abuses that are always harmful for the health.

In order to predict the risks of Alzheimer's Disease (AD) based on the deep learning model of brain 18F-FDG positron emission tomography (PET), a total of 350 mild cognitive impairment (MCI) participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database were selected as the research objects; in addition, the Convolutional Architecture for Fast Feature Embedding (CAFFE) was selected as the framework of the deep learning platform; the FDG PET image features of each participant were extracted by a deep convolution network model to construct the prediction and classification models; therefore, the MCI stage features were classified and the transformation was predicted. The results showed that in terms of the MCI transformation prediction, the sensitivity and specificity of conv3 classification were respectively 91.02% and 77.63%; in terms of the Late Mild Cognitive Impairment (LMCI) and Early Mild Cognitive Impairment (EMCI) classification, the accuracy of conv5 classification was 72.19%, and the sensitivity and specificity of conv5 were all 73% approximately. Thus, it was seen that the model constructed in the research could be used to solve the problems of MCI transformation prediction, which also had certain effects on the classifications of EMCI and LMCI. The risk prediction of AD based on the deep learning model of brain 18F-FDG PET discussed in the research matched the expected results. It provided a relatively accurate reference model for the prediction of AD. Despite the deficiencies of the research process, the research results have provided certain references and guidance for the future exploration of accurate AD prediction model; therefore, the research is of great significance. [ABSTRACT FROM AUTHOR]

Copyright of Quaderni Di Teoria Sociale is the property of Morlacchi Editore and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Based on the growing demand for neural network technologies, various neural network inference engines are being developed. However, each inference engine has its own neural network storage format. There is a growing demand for standardization to solve this problem. This study presents interworking techniques for ensuring the compatibility of neural networks and data among the various deep learning frameworks. The proposed technique standardizes the graphic expression grammar and learning data storage format using the Neural Network Exchange Format (NNEF) of Khronos. The proposed converter includes a lexical, syntax, and parser. This NNEF parser converts neural network information into a parsing tree and quantizes data. To validate the proposed system, we verified that MNIST is immediately executed by importing AlexNet's neural network and learned data. Therefore, this study contributes an efficient design technique for a converter that can execute a neural network and learned data in various frameworks regardless of the storage format of each framework. [ABSTRACT FROM AUTHOR]

Since the birth of artificial intelligence, the theory and the technology have become more mature, and the application field is expanding. Mobile networks and applications have grown quickly in recent years, and mobile computing is the new computing paradigm for mobile networks. In this paper, we build an artificial intelligence platform for a mobile service, which supports deep learning frameworks such as TensorFlow and Caffe. We describe the overall architecture of the AI platform for a GPU cluster in mobile service computing. In the GPU cluster, based on the scheduling layer, we propose Yarn by the Slurm scheduler to not only improve the distributed TensorFlow plug-in for the Slurm scheduling layer but also to extend YARN to manage and schedule GPUs. The front-end of the high-performance AI platform has the attributes of availability, scalability and efficiency. Finally, we verify the convenience, scalability, and effectiveness of the AI platform by comparing the performance of single-chip and distributed versions for the TensorFlow, Caffe and YARN systems. [ABSTRACT FROM AUTHOR]

In spite of the existence of numerous navigation tools/systems, Google Street View, offering only a single static image at a time, is still sometimes preferred for the provision of a realistic scene. However, for the sake of navigation, given the starting and ending locations, a navigation video consisting of images obtained from Google Street View service is desired. Several papers have tried to address this issue in some sense; however, there is still much room for further improvement. First, the generation of navigation video is not very smooth, i.e., the transition from one frame to another frame is not properly controlled, thus resulting a potential abrupt change from one scene toward another. Second, the generated video oftentimes contains many undesired vehicles and people, and the removal of these distracting objects would greatly enhance the quality of the navigational video. In this paper, we first make use of HOG and/or Haar features for detecting vehicles and people, and then we have also made some preliminary trials of using Faster R-CNN and Caffe to speed up detecting vehicles and people. Results are demonstrated to prove the effectiveness of our approaches and compared with similar approaches when applicable to show our improvement. In addition, a post-processing tool is also developed to interactively refine the results in case the automatic object detection is not perfect. [ABSTRACT FROM AUTHOR]

Deep neural network (DNN) has a deep hierarchy that connects multiple internal layers for feature detection and recognition. In our previous work, we proposed a vegetable recognition system which was based on Caffe framework. In this paper, we propose a vegetable category recognition system using DNN frameworks. We present a Vegeshop tool and website for users. Our system can be accessed ubiquitously from anywhere. We evaluate the performance of our vegetable category recognition using 15 kind of vegetables. Also, we evaluate the performance of learning accuracy and loss for vegetable recognition system which is based on Caffe and Chainer frameworks. In addition, we present the performance of recognition rate for different vegetables with different pixel sizes. The evaluation results show that the learning rate is more than 80%. We noticed that the performance of this recognition system is degraded when the color of the object is yellow. In this case, our system does not recognize the outline of the object by light intensity. From these studies, we found that the results of Caffe are higher than Chainer. For both frameworks, when pixel sizes is 256 × 256 , the results of accuracy is increased rapidly with the increase in iterations. [ABSTRACT FROM AUTHOR]

Deep learning techniques based on Convolutional Neural Networks (CNNs) are extensively used for the classification of hyperspectral images. These techniques present high computational cost. In this paper, a GPU (Graphics Processing Unit) implementation of a spatial-spectral supervised classification scheme based on CNNs and applied to remote sensing datasets is presented. In particular, two deep learning libraries, Caffe and CuDNN, are used and compared. In order to achieve an efficient GPU projection, different techniques and optimizations have been applied. The implemented scheme comprises Principal Component Analysis (PCA) to extract the main features, a patch extraction around each pixel to take the spatial information into account, one convolutional layer for processing the spectral information, and fully connected layers to perform the classification. To improve the initial GPU implementation accuracy, a second convolutional layer has been added. High speedups are obtained together with competitive classification accuracies. [ABSTRACT FROM AUTHOR]

Although 1990s Brazilian cinema revisits the sertão to find Brazilian identity, by the 2000’s some films begin challenging that narrative. As the state and its narratives move to the city, the sertão begins to be represented as a wasteland. This paper examines one such film, Narradores de Javé (The Storytellers; Caffé, 2003). Previously studied for its use of narration to countering the discourses of modernity, I propose that the film also constructs the wasteland as a social space by enacting affectivity to build a sense of communal narrative. First, I engage with Zygmunt Bauman’s Wasted Lives (2004) to explore how the state denies its inhabitants citizenship as subjectification. Second, in response to this disavowal, the villagers mobilize what Kathleen Stewart calls Ordinary Affects (2007) to subjectify by creating a social space through the communal sharing of stories that wasted objects allow them to recall.

Copyright of Journal of Zhengzhou University (Natural Science Edition) is the property of Journal of Zhengzhou University (Natural Science Edition) Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Deep learning techniques can automatically learn features from a large number of image data set. Automatic vegetable image classification is the base of many applications. This paper proposed a high performance method for vegetable images classification based on deep learning framework. The AlexNet network model in Caffe was used to train the vegetable image data set. The vegetable image data set was obtained from ImageNet and divided into training data set and test data set. The output function of the AlexNet network adopted the Rectified Linear Units (ReLU) instead of the traditional sigmoid function and the tanh function, which can speed up the training of the deep learning network. The dropout technology was used to improve the generalization of the model. The image data extension method was used to reduce overfitting in the learning process. With AlexNet network model used for training different number of vegetable image data set, the experimental results showed that the classification accuracy decreases as the number of data set decreases. The experimental verification indicated that the accuracy rate of the deep learning method in the test data set reached as high as 92.1%, which was greatly improved compared with BP neural network (78%) and SVM classifier (80.5%) methods. [ABSTRACT FROM AUTHOR]

Auscultation is one of the most used techniques for detecting cardiovascular diseases, which is one of the main causes of death in the world. Heart murmurs are the most common abnormal finding when a patient visits the physician for auscultation. These heart sounds can either be innocent, which are harmless, or abnormal, which may be a sign of a more serious heart condition. However, the accuracy rate of primary care physicians and expert cardiologists when auscultating is not good enough to avoid most of both type-I (healthy patients are sent for echocardiogram) and type-II (pathological patients are sent home without medication or treatment) errors made. In this paper, the authors present a novel convolutional neural network based tool for classifying between healthy people and pathological patients using a neuromorphic auditory sensor for FPGA that is able to decompose the audio into frequency bands in real time. For this purpose, different networks have been trained with the heart murmur information contained in heart sound recordings obtained from nine different heart sound databases sourced from multiple research groups. These samples are segmented and preprocessed using the neuromorphic auditory sensor to decompose their audio information into frequency bands and, after that, sonogram images with the same size are generated. These images have been used to train and test different convolutional neural network architectures. The best results have been obtained with a modified version of the AlexNet model, achieving 97% accuracy (specificity: 95.12%, sensitivity: 93.20%, PhysioNet/CinC Challenge 2016 score: 0.9416). This tool could aid cardiologists and primary care physicians in the auscultation process, improving the decision making task and reducing type-I and type-II errors. [ABSTRACT FROM PUBLISHER]

Until the 16th century, the European words for 'coffeeʼ reflect exclusively the Arabic and Turkish type qahwa / kahve. The first Italian - and at the same time the earliest European - record appears in 1579 ( cava) in the Viaggio da Creta in Egitto ed al Sinai of Filippo Pigafetta. From the 17th century onwards, the type It. caffè, Fr. café, Germ. Kaffee, Engl. coffee, which today is predominant all over Europe, begins to make its way. Phonetically, the genesis of the voiceless variants in Western Europe is possible on the basis of both the Turkish and the Arabic model. [ABSTRACT FROM AUTHOR]

We propose a practical Convolution Neural Network (CNN) model termed the CNN for Semantic Segmentation for driver Assistance system (CSSA). It is a novel semantic segmentation model for probabilistic pixel-wise segmentation, which is able to predict pixel-wise class labels of a given input image. Recently, scene understanding has turned out to be one of the emerging areas of research, and pixel-wise semantic segmentation is a key tool for visual scene understanding. Among future intelligent systems, the Advanced Driver Assistance System (ADAS) is one of the most favorite research topic. The CSSA is a road scene understanding CNN that could be a useful constituent of the ADAS toolkit. The proposed CNN network is an encoder-decoder model, which is built on convolutional encoder layers adopted from the Visual Geometry Group's VGG-16 net, whereas the decoder is inspired by segmentation network (SegNet). The proposed architecture mitigates the limitations of the existing methods based on state-of-the-art encoder-decoder design. The encoder performs convolution, while the decoder is responsible for deconvolution and un-pooling/up-sampling to predict pixel-wise class labels. The key idea is to apply the up-sampling decoder network, which maps the low-resolution encoder feature maps. This architecture substantially reduces the number of trainable parameters and reuses the encoder's pooling indices to up-sample to map pixel-wise classification and segmentation. We have experimented with different activation functions, pooling methods, dropout units and architectures to design an efficient CNN architecture. The proposed network offers a significant improvement in performance in segmentation results while reducing the number of trainable parameters. Moreover, there is a considerable improvement in performance in comparison to the benchmark results over PASCAL VOC-12 and the CamVid. [ABSTRACT FROM AUTHOR]