Emotion recognition systems have been used frequently both socially and commercially. Various classification methods are used to perform emotion recognition. The convolutional neural network (CNN) has a popular positi...
详细信息
ISBN:
(纸本)9781728119045
Emotion recognition systems have been used frequently both socially and commercially. Various classification methods are used to perform emotion recognition. The convolutional neural network (CNN) has a popular position among these classification methods. The CNN modeling process, which is left to the user experience, is a hard challenging task. Instead of modeling a new system, transfer learning from pre-trained models often increases performance and prevents loss of time. In this paper, a new dataset, created with the help of image search engines, is classified with transfer learning supported CNN. Data preprocessingmethods were also applied to increase classification success. The results obtained with experimental studies are presented in detail.
Fast and small-resourced implementation of convolutional neural network (CNN) into a field-programmable gate array (FPGA) was realized using a binarized neural network (NN). We propose a set of neuron and network mode...
详细信息
ISBN:
(纸本)9781728127453
Fast and small-resourced implementation of convolutional neural network (CNN) into a field-programmable gate array (FPGA) was realized using a binarized neural network (NN). We propose a set of neuron and network models optimized for fully binarized implementation of general NNs using the look-up-tables (LUTs) in modern FPGAs, which is herein referred to as sparse-LUT model. Arrayed MNIST data images of more than 40 characters input from a camera were recognized with 92.8% accuracy and classified with the colored marks on the characters on organic light-emitting diode (OLED) display images with 1-ms cycle time, <1.0-ms delay in the LUT-based CNN recognition, and <2-ms total time delay. In combination with stochastic time-divided signalprocessing, binarized signals in this model can be extended for processing multi-bit (analogue-like) signals in an oversampling manner with increased recognition accuracies up to 98.6% in MNIST and 58.3% in CIFAR-10 image data sets using CNN. The source codes for the binarized NN core were released and open-sourced.
Deep brain stimulation (DBS) is an interventional treatment for Parkinson's disease which involves the precise positioning of stimulated electrodes within deep brain structures, such as the Subthalamic Nucleus (ST...
详细信息
Deep brain stimulation (DBS) is an interventional treatment for Parkinson's disease which involves the precise positioning of stimulated electrodes within deep brain structures, such as the Subthalamic Nucleus (STN). Although originally identified via imaging, additional inter-operative guidance is necessary to localize the target anatomy. Analysis of Micro-Electrode Recordings (MERs) allows for a trained neurophysiologist to infer the underlying anatomy at a particular electrode position using human audition, although it is subjective and requires a high degree of expertise. Various approaches to assist MER analysis during DBS are proposed in the literature, including deep learning methods, which rely on a static input description, that is, a pre-defined number of features or input size. In this paper, we propose two dynamic deep learning approaches adaptable to the complexity of MERs signal, by using an arbitrary long listening time (in 1s chunks), while providing feedback to the neurophysiologist as to the model's certainty. We evaluated five different deep learning based classifiers which can use arbitrary length MERs for STN segmentation. We found that a Bayesian extension using the high-level features from SepaConvNet performed the best, increasing the balanced accuracy to 83.5%. This work represents a step forward in integrating automated analysis of MERs into the DBS surgical workflow by automatically finding and exploiting possible efficiencies in MER acquisition.
Recently deep learning has been introduced to the field of image compression. In this paper, we present a hybrid coding framework that combines entropy coding, deep learning, and traditional coding framework. In the b...
详细信息
Recently deep learning has been introduced to the field of image compression. In this paper, we present a hybrid coding framework that combines entropy coding, deep learning, and traditional coding framework. In the base layer of the encoding, we use convolutional neural networks to lear n the latent representation and importance map of the original image respectively. The importance map is then used to guide the bit allocation of the latent representation . A context model is also developed to help the entropy coding after the masked quantization. Another network is used to get a coarse reconstruction of the image in the base layer. The residual between the input and the coarse reconstruction is then obtained and encoded by the traditional BPG codec as the enhancement layer of the bit stream. We only need to train a basic model and the proposed scheme can realize image compression at different bit rates, thanks to the use of the traditional codec. Experimental results using the Kodak, Urban100 and BSD100 datasets show that the proposed scheme outperforms many deep learning-based methods and traditional codecs including BPG in MS-SSIM metric across a wide range of bit rates. It also exceeds some latest hybrid schemes in RGB444 domain on Kodak dataset in both PSN R and MS-SSIM metrics.
Recognition of human action is one of the challenges in the field of artificial intelligence. Deep learning model has become a research issue in action recognition applications due to its ability to outperform traditi...
详细信息
Context. The availability of large bandwidth receivers for millimeter radio telescopes allows for the acquisition of position-position-frequency data cubes over a wide field of view and a broad frequency coverage. The...
详细信息
Context. The availability of large bandwidth receivers for millimeter radio telescopes allows for the acquisition of position-position-frequency data cubes over a wide field of view and a broad frequency coverage. These cubes contain a lot of information on the physical, chemical, and kinematical properties of the emitting gas. However, their large size coupled with an inhomogenous signal-to-noise ratio (S/N) are major challenges for consistent analysis and ***. We searched for a denoising method of the low S/N regions of the studied data cubes that would allow the low S/N emission to be recovered without distorting the signals with a high S/***. We performed an in-depth data analysis of the (CO)-C-13 and (CO)-O-17 (1-0) data cubes obtained as part of the ORION-B large program performed at the IRAM 30 m telescope. We analyzed the statistical properties of the noise and the evolution of the correlation of the signal in a given frequency channel with that of the adjacent channels. This has allowed us to propose significant improvements of typical autoassociative neural networks, often used to denoise hyperspectral Earth remote sensing data. Applying this method to the (CO)-C-13 (1-0) cube, we were able to compare the denoised data with those derived with the multiple Gaussian fitting algorithm ROHSA, considered as the state-of-the-art procedure for data line ***. The nature of astronomical spectral data cubes is distinct from that of the hyperspectral data usually studied in the Earth remote sensing literature because the observed intensities become statistically independent beyond a short channel separation. This lack of redundancy in data has led us to adapt the method, notably by taking into account the sparsity of the signal along the spectral axis. The application of the proposed algorithm leads to an increase in the S/N in voxels with a weak signal, while preserving the spectral shape of the data in high S/N ***. Th
Warm restart strategies are widely used in gradient-free optimization to deal with multi-model functions. In this paper, we present a novel warm restart technique by step cosine function in stochastic gradient descent...
详细信息
ISBN:
(数字)9781728168968
ISBN:
(纸本)9781728168975
Warm restart strategies are widely used in gradient-free optimization to deal with multi-model functions. In this paper, we present a novel warm restart technique by step cosine function in stochastic gradient descent method that used to train a deep convolution neural network. Three variants of step cosine function with MobileNetv2 and ResNet50 network structure are tested in our pathological lymph node PET/CT dataset. Comparing to the step function as the warm restart schedule, the proposed step cosine warm restart strategy could improve the performance of pathological lymph node image classification in terms of accuracy, sensitivity and specificity, which increased at 2.1%, 0.7% and 2.9% with MobileNetv2, and at 1.3%, 1.4% and 1.3% with ResNet50, respectively.
Various methods have been proposed for the automatic speech recognition system. The main emphasis has been given to the signal or voice classification of individuals. In this paper, we are considering the four promine...
Various methods have been proposed for the automatic speech recognition system. The main emphasis has been given to the signal or voice classification of individuals. In this paper, we are considering the four prominent neural network architectures namely, Multilayer Feed-Forward neural Network, Cascade Backpropagation Network, Elman Backpropagation Network and Adaptive Linear neural Network for the performance evaluation on collected sound samples of ten different people. These networks have been trained using mini-batch stochastic gradient descent learning with batch normalization. The feed-forward neural network model has been extended up to the four hidden layers to perform the desired classification of sound samples. The collected samples of the analog form of sound signals are converted into the digital form using the digital signalprocessing technique. The time & frequency patches of these digital signals have been considered as the training pattern samples. The test pattern sets have also been constructed in the same manner and the performances of these neural network architectures have been measured. The simulation results are exhibiting the better performance of Cascade neural Network over the other architectures.
暂无评论