|
|
|
题名
|
作者
|
年代
|
出处
|
被引量
|
| 1 | Achieving Safe Deep Reinforcement Learning via Environment Comprehension Mechanism显示文摘Deep reinforcement learning(DRL), which combines deep learning with reinforcement learning, has achieved great success recently. In some cases, however,during the learning process agents may reach states that are worthless and dangerous where the task fails. To address the problem, we propose an algorithm, referred as Environment comprehension mechanism(ECM) for deep reinforcement learning to attain safer decisions. ECM perceives hidden dangerous situations by analyzing object and comprehending the environment, such that the agent bypasses inappropriate actions systematically by setting up constraints dynamically according to states. ECM, which calculates the gradient of the states in Markov tuple, sets up boundary conditions and generates a rule to control the direction of the agent to skip unsafe states. ECM is able to be applied to basic deep reinforcement learning algorithms to guide the selection of actions. The experiment results show that the algorithm promoted safety and stability of the control tasks. | PENG Pai ZHU Fei LIU Quan ZHAO Peiyao WU Wen | 2021 | Chinese Journal of Electronics2021，30，6： | 2 |
| 2 | Words in Pairs Neural Networks for Text Classification显示文摘Existing methods utilized single words as text features.Some words contain multiple meanings,and it is difficult to distinguish its specific classification according to a single word,which probably affects the accuracy of the text classification.Propose a framework based on Words in pairs neural networks(WPNN)for text classification.Words in pairs include all single word combinations which have a high mutual association.Mine the crucial explicit and implicit Words in pairs as text features.These words in pairs as a text feature are easily classified.The words in pairs are utilized as the input of the neural network,which provides a better classification ability to the model,because they are more recognizable than the single word.Experimental results show that our model outperforms five benchmark algorithms. | WU Yujia LI Jing SONG Chengfang CHANG Jun | 2020 | Chinese Journal of Electronics2020，29，3： | 1 |
| 3 | Porn Streamer Recognition in Live Video Based on Multimodal Knowledge Distillation显示文摘Although deep learning has reached a higher accuracy for video content analysis,it is not satisfied with practical application demands of porn streamer recognition in live video because of multiple parameters,complex structures of deep network model.In order to improve the recognition efficiency of porn streamer in live video,a deep network model compression method based on multimodal knowledge distillation is proposed.First,the teacher model is trained with visual-speech deep network to obtain the corresponding porn video prediction score.Second,a lightweight student model constructed with Mobile Net V2 and Xception transfers the knowledge from the teacher model by using multimodal knowledge distillation strategy.Finally,porn streamer in live video is recognized by combining the lightweight student model of visualspeech network with the bullet screen text recognition network.Experimental results demonstrate that the proposed method can effectively drop the computation cost and improve the recognition speed under the proper accuracy. | WANG Liyuan ZHANG Jing YAO Jiacheng ZHUO Li | 2021 | Chinese Journal of Electronics2021，30，6： | 1 |
| 4 | A Hierarchical Scheme for Video-Based Person Re-identification Using Lightweight PCANet and Handcrafted LOMO Features显示文摘A two-level hierarchical scheme for video-based person re-identification(re-id)is presented,with the aim of learning a pedestrian appearance model through more complete walking cycle extraction.Specifically,given a video with consecutive frames,the objective of the first level is to detect the key frame with lightweight Convolutional neural network(CNN)of PCANet to reflect the summary of the video content.At the second level,on the basis of the detected key frame,the pedestrian walking cycle is extracted from the long video sequence.Moreover,local features of Local maximal occurrence(LOMO)of the walking cycle are extracted to represent the pedestrian’s appearance information.In contrast to the existing walking-cycle-based person re-id approaches,the proposed scheme relaxes the limit on step number for a walking cycle,thus making it flexible and less affected by noisy frames.Experiments are conducted on two benchmark datasets:PRID 2011 and i LIDS-VID.The experimental results demonstrate that our proposed scheme outperforms the six state-of-art video-based re-id methods,and is more robust to the severe video noises and variations in pose,lighting,and camera viewpoint. | LI Youjiao ZHUO Li LI Jiafeng ZHANG Jing | 2021 | Chinese Journal of Electronics2021，30，2： | 0 |
您所在的机构暂未开通此部分的功能，请联系管理员