Spectrum-induced polarization (SIP) is a widely used geophysical exploration approach, but it is prone to noise. To address this issue, this paper proposes a noise reduction method that combines phase space reconstruc...
详细信息
In the cement production process, the decomposer outlet temperature control faces challenges such as large time delay and multiple disturbances. A general controlsystem cannot handle time delay and disturbances effec...
详细信息
In the cement production process, the speed of grate coolers directly affects the residence time and cooling effect of cement clinker. Its effective control is of great significance to ensuring cement quality. But the...
详细信息
Augmented reality technology is a research hotspot in the field of computer vision in recent years. Tracking registration technology determines the performance of augmented reality systems. The accuracy of tracking re...
详细信息
An improved method for spectral reflectance reconstruction from the RGB response of the digital camera is proposed by deep convolution neural network. The proposed method learns a fusion mapping theory that represents...
详细信息
For speech emotion recognition (SER), emotional feature set with high dimension may produce redundant features and influence the recognition rate. To solve this problem, feature selection of speaker-independent speech...
详细信息
For speech emotion recognition (SER), emotional feature set with high dimension may produce redundant features and influence the recognition rate. To solve this problem, feature selection of speaker-independent speech based on genetic algorithm (GA) is proposed, which can obtain optimal feature subset. And a four-level emotional classification method based on support vector machine (SVM) is proposed according to the confusion degree among different emotional categories. A framework of speaker-independent SER is presented and the classification experiments based on proposed methods by using Chinese speech database from institute of automation of Chinese academy of sciences (CASIA) are performed, where the speaker-independent features selected by the proposed feature selection method and Spearman correlation analysis are used for emotion recognition, respectively. The experimental results show that the proposal achieved 77.6% recognition rate on average, which is about 1.2% higher than other recognition methods. By proposal, it would be efficient to distinguish the emotional states of different speakers from speech.
In order to improve the accuracy of short-term power load forecasting, an algorithm that simultaneously optimizes SVR parameters and optimizes the selection of input features is proposed. Firstly, the input features a...
详细信息
In this paper, the fault detection problem has been investigated for networked singularly perturbed systems with time delays under the stochastic communication protocol (SCP). To avoid/alleviate the undesired data col...
详细信息
High accurate rate of street objects detection is significant to realize intelligent vehicles. Algorithms based on Convolution Neural Network (CNN) have already shown their reasonable performance on general object det...
详细信息
High accurate rate of street objects detection is significant to realize intelligent vehicles. Algorithms based on Convolution Neural Network (CNN) have already shown their reasonable performance on general object detection. For example SSD and YOLO can detection wide variety of objects on 2D images in real time, but the performance is not good enough on street objects detection especially on complex urban street environment. In this paper, instead of proposing and training a new CNN model, we use transfer learning methods to learn from generic CNN model to our specific model to achieve good performance. The transfer learning methods include fine-tuning the pretrained CNN model with self-made dataset and adjusting CNN model structure. We analyze transfer learning results on fine-tuning Single shot multibox detector (SSD) with self-made datasets. The experimental results based on transfer learning method show that the proposed method is effective.
Compared with speech, facial expression, and body languages, Electroencephalogram (EEG) can reflect the inner activity of brain, by which the emotion can be recognized objectively and naturally. In this paper, an EEG ...
详细信息
Compared with speech, facial expression, and body languages, Electroencephalogram (EEG) can reflect the inner activity of brain, by which the emotion can be recognized objectively and naturally. In this paper, an EEG emotion recognition system is proposed in which EEG signals of 6 channels are detected from Frontal Lobe and Temporal Lobe, and then the time-domain features of statistics features and frequency-domain features of spectrum centroid (SC) are extracted. To remove the redundant feature, Linear Discriminant Analysis (LDA) is used to reduce the dimension of feature. In addition, an improved classifier based on PSO-SVM is applied to classify the emotional states in the Valance-Arousal emotion model, respectively, which are defined as High-Valance (HV) and Low-Valance (LV) on the Valance dimension and High-Arousal (HA) and Low-Arousal (LA) on the Arousal dimension. EEG emotion recognition experiment on DEAP dataset is performed, from which the results show that the proposed method obtains the accuracies of 73.33% on Valance dimension and 72.78% on Arousal dimension, which are higher than those of some state-of-the art works.
暂无评论