In recent years, driven by hardware technology, the computing power and programmability of GPUs have been rapidly developed. With the characteristics of highly parallel computing, GPUs are no longer limited to daily graphics processing tasks. It begins to involve a wider range of high-performance generalpurpose computing field. One of the hotspots in the field of highperformance parallel computing is MapReduce, a massive data processing framework. Through inexpensive ordinary computer clusters, we can obtain large-scale data computing capabilities that were previously only owned by expensive large servers. However, most existing MapReduce systems run on CPU clusters, and the computing performance of a single node is limited. Therefore, this paper proposes a parallel computing framework based on GPU cluster and MapReduce, and validates the effectiveness of the framework through experiments. Experiments have proven that our framework can complete the work, and it has a significant speedup for large-scale applications.
{"title":"Parallel Computing Framework Based on MapReduce and GPU Clusters","authors":"Chunlei Xu, Weijin Zhuang","doi":"10.1145/3424978.3425051","DOIUrl":"https://doi.org/10.1145/3424978.3425051","url":null,"abstract":"In recent years, driven by hardware technology, the computing power and programmability of GPUs have been rapidly developed. With the characteristics of highly parallel computing, GPUs are no longer limited to daily graphics processing tasks. It begins to involve a wider range of high-performance generalpurpose computing field. One of the hotspots in the field of highperformance parallel computing is MapReduce, a massive data processing framework. Through inexpensive ordinary computer clusters, we can obtain large-scale data computing capabilities that were previously only owned by expensive large servers. However, most existing MapReduce systems run on CPU clusters, and the computing performance of a single node is limited. Therefore, this paper proposes a parallel computing framework based on GPU cluster and MapReduce, and validates the effectiveness of the framework through experiments. Experiments have proven that our framework can complete the work, and it has a significant speedup for large-scale applications.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"248 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132856563","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The lower limb exoskeleton robot is a wearable device that enhances the human lower extremity movement ability. And gait phase detection is an important prerequisite for controlling the lower limb exoskeleton robot. Traditional gait phase detection is mostly based on ground contact forces (GCFs) measured by force sensitive resistors (FSRs). However, FSRs will lose its lifespan and accuracy due to the impact force generated by gait. In view of this shortcoming, a gait phase detection method based on the joints angle of lower limb is proposed. Stacked LSTMs was constructed by using joints angle information of lower limb exoskeleton as input and gait phase as output. Through the experimental analysis of the different wearers' gait phase detection results, Stacked LSTMs could effectively detect the gait phase through the joints angle information with an average accuracy rate of 94.1%, which has a certain role in simplifying the exoskeleton robot sensor network.
{"title":"Gait Phase Detection of Exoskeleton Robot Based on the Joints Angle of Lower Limb","authors":"Wang Jiang, Jianbin Zheng, Liping Huang","doi":"10.1145/3424978.3425067","DOIUrl":"https://doi.org/10.1145/3424978.3425067","url":null,"abstract":"The lower limb exoskeleton robot is a wearable device that enhances the human lower extremity movement ability. And gait phase detection is an important prerequisite for controlling the lower limb exoskeleton robot. Traditional gait phase detection is mostly based on ground contact forces (GCFs) measured by force sensitive resistors (FSRs). However, FSRs will lose its lifespan and accuracy due to the impact force generated by gait. In view of this shortcoming, a gait phase detection method based on the joints angle of lower limb is proposed. Stacked LSTMs was constructed by using joints angle information of lower limb exoskeleton as input and gait phase as output. Through the experimental analysis of the different wearers' gait phase detection results, Stacked LSTMs could effectively detect the gait phase through the joints angle information with an average accuracy rate of 94.1%, which has a certain role in simplifying the exoskeleton robot sensor network.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"50 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134526448","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
RF signals are widely used in space telemetry, track and command (TT&C) field. However, in the transmission process, a lot of noise will be introduced due to the interference of equipment components, transmission channel, atmosphere, electromagnetic environment, etc., which will affect the subsequent analysis and processing of the receiving equipment. Based on the singular value decomposition (SVD) method for noise suppression of RF signals, the Letts' criterion method was proposed to determine the effective rank order of singular value sequence (SVS). The effect of SVD on noise suppression in different dimension matrices were compared and analyzed. Main influencing factors were put forward to choose the matrix dimension as a result. Finally, Hankel matrix dimension automatic determination system was built to realize the choice of the matrix dimension. The noise suppression effect was improved by 0.5dB at least which compared with the traditional matrix dimension determination method.
{"title":"Determination Method of Effective Rank Degree and Matrix Dimension in SVD De-noising","authors":"Junyao Li, Yalong Yan, Weina Guo, Yangsongyi Su","doi":"10.1145/3424978.3425144","DOIUrl":"https://doi.org/10.1145/3424978.3425144","url":null,"abstract":"RF signals are widely used in space telemetry, track and command (TT&C) field. However, in the transmission process, a lot of noise will be introduced due to the interference of equipment components, transmission channel, atmosphere, electromagnetic environment, etc., which will affect the subsequent analysis and processing of the receiving equipment. Based on the singular value decomposition (SVD) method for noise suppression of RF signals, the Letts' criterion method was proposed to determine the effective rank order of singular value sequence (SVS). The effect of SVD on noise suppression in different dimension matrices were compared and analyzed. Main influencing factors were put forward to choose the matrix dimension as a result. Finally, Hankel matrix dimension automatic determination system was built to realize the choice of the matrix dimension. The noise suppression effect was improved by 0.5dB at least which compared with the traditional matrix dimension determination method.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"66 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133896646","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Different attributes and weights of decision-makers are the key information of multi-attribute and group decision making. Aiming at the complete unknown problem of attribute and decision-maker's weight information, this paper proposed a weight calculation method based on triangular intuitionistic fuzzy number, which established a consensus model with reference triangular intuitionistic fuzzy information by similarity degree to calculate attribute and decision-maker weight information. Aiming at the partially unknown problem of attribute and decision-maker's weight, based on the consensus model, linear programming method is used to calculate the attribute and decision-maker's weight information. Combined with the method of TODIM decision based on triangular intuitionistic fuzzy information, it can be applied to scientific decision making in modern enterprise management in many application scenarios, such as known, partially unknown and completely unknown attribute and decision-maker weight.
{"title":"Research on Multi-attribute and Group Decision-making Method with Unknown Weight","authors":"Yang Xie, Gongliang Li, Qingfei Cai","doi":"10.1145/3424978.3425148","DOIUrl":"https://doi.org/10.1145/3424978.3425148","url":null,"abstract":"Different attributes and weights of decision-makers are the key information of multi-attribute and group decision making. Aiming at the complete unknown problem of attribute and decision-maker's weight information, this paper proposed a weight calculation method based on triangular intuitionistic fuzzy number, which established a consensus model with reference triangular intuitionistic fuzzy information by similarity degree to calculate attribute and decision-maker weight information. Aiming at the partially unknown problem of attribute and decision-maker's weight, based on the consensus model, linear programming method is used to calculate the attribute and decision-maker's weight information. Combined with the method of TODIM decision based on triangular intuitionistic fuzzy information, it can be applied to scientific decision making in modern enterprise management in many application scenarios, such as known, partially unknown and completely unknown attribute and decision-maker weight.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133546009","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Accurate gait phase recognition and gait cycle segmentation are the basis for analyzing individual gait. This paper introduces a ground reaction force (GRF) signal analysis method using a portable, wearable gait analysis system. In this paper, we make use of the signal obtained from the 8 pressure sensors, and use fuzzy logic inference to achieve continuous and smooth gait phase recognition. Then, gait cycle segmentation is performed using gait phases by fully considering the internal difference among different people. The proposed gait segmentation algorithm does not need to preset the phase sequence that forms the individual gait, which can detect accurate gait patterns regardless of the users. Experimental results show that the proposed algorithm has 97.2% accuracy that is similar to the traditional gait cycle segmentation method based on the empirical formula.
{"title":"Research on Gait Cycle Recognition with Plantar Pressure Sensors","authors":"Yina Yang, Weidong Gao, Zhenwei Zhao","doi":"10.1145/3424978.3424998","DOIUrl":"https://doi.org/10.1145/3424978.3424998","url":null,"abstract":"Accurate gait phase recognition and gait cycle segmentation are the basis for analyzing individual gait. This paper introduces a ground reaction force (GRF) signal analysis method using a portable, wearable gait analysis system. In this paper, we make use of the signal obtained from the 8 pressure sensors, and use fuzzy logic inference to achieve continuous and smooth gait phase recognition. Then, gait cycle segmentation is performed using gait phases by fully considering the internal difference among different people. The proposed gait segmentation algorithm does not need to preset the phase sequence that forms the individual gait, which can detect accurate gait patterns regardless of the users. Experimental results show that the proposed algorithm has 97.2% accuracy that is similar to the traditional gait cycle segmentation method based on the empirical formula.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133675560","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
With the wide application of X-ray screening machines, the intelligent recognition of contrabands in the X-ray screening images has been paid more and more attention. Contrabands detection in X-ray screening images is a challenging problem in the field of security detection due to the random distribution of the items, which can cause the overlapping the target objects and the other objects. It is difficult to segment the X-ray security images into the different candidate regions which contain different objects by traditional image processing and recognition algorithm. In recent years, YOLO (You only look once, a realtime object detection system) Model was presented which provides a simple framework to predict bounding boxes and class probabilities directly from full images. In this paper, a YOLO based model is used to detect the contrabands in X-ray screening images. The experimental results show that the precision and the recall rate of contrabands detection under simple background are respectively higher than 98 percent and 94 percent. In complex environment, the precision remains above 95 percent, but the recall rate of some kinds of contrabands dropped down to about 70 percent.
随着x射线筛检机的广泛应用，对x射线筛检图像中违禁品的智能识别越来越受到重视。x射线扫描图像中的违禁品检测是安全检测领域的一个难题，因为违禁品的随机分布会导致目标物体与其他物体重叠。传统的图像处理和识别算法难以将x射线安全图像分割成包含不同目标的不同候选区域。近年来提出了YOLO (You only look once，实时目标检测系统)模型，该模型提供了一个简单的框架，可以直接从完整图像中预测边界框和类别概率。本文采用基于YOLO的模型对x射线扫描图像中的违禁品进行检测。实验结果表明，在简单背景下，违禁品检测的准确率和召回率分别高于98%和94%。在复杂的环境下，准确率保持在95%以上，但某些违禁品的召回率下降到70%左右。
{"title":"Contrabands Detection in X-ray Screening Images Using YOLO Model","authors":"Ju Wu, Huan Shi, Qinxue Wang","doi":"10.1145/3424978.3425106","DOIUrl":"https://doi.org/10.1145/3424978.3425106","url":null,"abstract":"With the wide application of X-ray screening machines, the intelligent recognition of contrabands in the X-ray screening images has been paid more and more attention. Contrabands detection in X-ray screening images is a challenging problem in the field of security detection due to the random distribution of the items, which can cause the overlapping the target objects and the other objects. It is difficult to segment the X-ray security images into the different candidate regions which contain different objects by traditional image processing and recognition algorithm. In recent years, YOLO (You only look once, a realtime object detection system) Model was presented which provides a simple framework to predict bounding boxes and class probabilities directly from full images. In this paper, a YOLO based model is used to detect the contrabands in X-ray screening images. The experimental results show that the precision and the recall rate of contrabands detection under simple background are respectively higher than 98 percent and 94 percent. In complex environment, the precision remains above 95 percent, but the recall rate of some kinds of contrabands dropped down to about 70 percent.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114213508","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Intelligent autonomous control and intelligent information application are the main features of an intelligent launch site. The features put forward high demands for the construction of the control system in ground support facilities. The paper analyzes the main characteristics of the ground support control system in the intelligent launch site and the corresponding crucial technical development as well as its application. The construction of ground support control system involved in the crucial technologies that mainly have three aspects: data acquisition and processing, system application and system assessment. The architecture of ground support control system is proposed as a three-dimensional general framework in an intelligent launch site. The architecture is designed by five layers including equipment level, platform level, algorithm level, management level, and application level. Based on the architecture, the operation mechanism of the ground support control systems is presented.
{"title":"Research on the Construction of Control System in Ground Support Facilities at Intelligent Launch Site","authors":"Wei Dong, Litian Xiao, Gang Lei, Zhaojian Li, Fenglei Zu, Wenyi Zhuang","doi":"10.1145/3424978.3425055","DOIUrl":"https://doi.org/10.1145/3424978.3425055","url":null,"abstract":"Intelligent autonomous control and intelligent information application are the main features of an intelligent launch site. The features put forward high demands for the construction of the control system in ground support facilities. The paper analyzes the main characteristics of the ground support control system in the intelligent launch site and the corresponding crucial technical development as well as its application. The construction of ground support control system involved in the crucial technologies that mainly have three aspects: data acquisition and processing, system application and system assessment. The architecture of ground support control system is proposed as a three-dimensional general framework in an intelligent launch site. The architecture is designed by five layers including equipment level, platform level, algorithm level, management level, and application level. Based on the architecture, the operation mechanism of the ground support control systems is presented.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116453906","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Aiming at the problem of prior knowledge acquisition in the process of Bayesian network construction, AHP/D-S evidence theory is introduced into Bayesian network parameter learning. An algorithm that uses AHP/D-S evidence theory to integrate expert prior knowledge, integrates monotonic constraints and near-equal constraints for parameter learning is proposed, and simulation cases are studied. Given corrective expert prior knowledge, the new parameter-learning algorithm overcomes the shortcomings of miscalculation and miscalculation of certain small probability parameters under the condition of small sample set by MLE, and was obviously better than MLE and MAP without prior information. This paper provides a new method for acquiring prior knowledge in the Bayesian network parameter learning process.
{"title":"Bayesian Network Parameter Learning Method Based on AHP/D-S Evidence Theory","authors":"Shuhuan Wei, Yanqiao Chen, Junbao Geng","doi":"10.1145/3424978.3425149","DOIUrl":"https://doi.org/10.1145/3424978.3425149","url":null,"abstract":"Aiming at the problem of prior knowledge acquisition in the process of Bayesian network construction, AHP/D-S evidence theory is introduced into Bayesian network parameter learning. An algorithm that uses AHP/D-S evidence theory to integrate expert prior knowledge, integrates monotonic constraints and near-equal constraints for parameter learning is proposed, and simulation cases are studied. Given corrective expert prior knowledge, the new parameter-learning algorithm overcomes the shortcomings of miscalculation and miscalculation of certain small probability parameters under the condition of small sample set by MLE, and was obviously better than MLE and MAP without prior information. This paper provides a new method for acquiring prior knowledge in the Bayesian network parameter learning process.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"366 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132273376","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
According to the time varying of Received Signal Strength (RSS) and the difference of received signal capability among different terminals, which leads to the instability and inaccuracy of Wi-Fi indoor positioning, a novel Wi-Fi positioning method based on RSS matrix correlation is proposed. This method firstly collects Wi-Fi fingerprint data of all reference points in the off-line training stage, constructs an RSS matrix by filtering and sorting the fingerprint data, and records the coordinates of the reference points and the corresponding RSS matrix to establish the off-line location fingerprint database. In the positioning stage, by comparing the RSS matrix correlation between the real-time monitoring and the reference point in the off-line fingerprint database to find the most relevant k reference points, and then estimate the final position of user by weighting centroid algorithm. Experimental results show that this method has better positioning accuracy than the traditional indoor Wi-Fi positioning, and reduce the impact of different terminals on indoor positioning, thus improving the stability of positioning.
{"title":"An Indoor Wi-Fi Positioning Method Based on RSS Matrix Relevance","authors":"Tao Zheng, Guanping Hua, B. Zhu","doi":"10.1145/3424978.3424983","DOIUrl":"https://doi.org/10.1145/3424978.3424983","url":null,"abstract":"According to the time varying of Received Signal Strength (RSS) and the difference of received signal capability among different terminals, which leads to the instability and inaccuracy of Wi-Fi indoor positioning, a novel Wi-Fi positioning method based on RSS matrix correlation is proposed. This method firstly collects Wi-Fi fingerprint data of all reference points in the off-line training stage, constructs an RSS matrix by filtering and sorting the fingerprint data, and records the coordinates of the reference points and the corresponding RSS matrix to establish the off-line location fingerprint database. In the positioning stage, by comparing the RSS matrix correlation between the real-time monitoring and the reference point in the off-line fingerprint database to find the most relevant k reference points, and then estimate the final position of user by weighting centroid algorithm. Experimental results show that this method has better positioning accuracy than the traditional indoor Wi-Fi positioning, and reduce the impact of different terminals on indoor positioning, thus improving the stability of positioning.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"49 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128206910","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
For agricultural special species, the labeled procedure of large-scale samples is costly, thus, the bamboo species only has a limited number for supervised learning. The fine-tuning strategy is important for deep neural network by transferring learning methods, which utilize the weight of the deep model of the source domain, and can solve the problem associated with insufficient samples to make the model more stability and robustness. In the manuscript, the novelty of the strategy, for images of bamboo species with low-shot classification, mainly proposed an idea that is the transfer of the convolutional group features of deep convolutional models. The deep models with a novel fine-tuning method and three optimizers that are stochastic gradient descent, Adaptive Moment estimation, and Adadelta respectively, are evaluated by the accuracy and the expected calibration error value for the analysis of deep model generalization. An analysis of the results showed that, based on the proportion of training dataset is only 30%, the innovative strategy for bamboo species classification achieved better performance that has an accuracy of 0.82, and the expected calibration error of 0.16, which were better stability and generalization than those of other fine-tuning strategies. Consequently, the novel fine-tuning strategy proposed in this manuscript transfers the features of deep convolutional groups, improves the accuracy and generalizability of the model, and resolves the problems associated with having insufficient samples of bamboo species for low-shot classification.
{"title":"Classifying a Limited Number of the Bamboo Species by the Transformation of Convolution Groups","authors":"Xiu Jin, Xianzhi Zhu","doi":"10.1145/3424978.3425009","DOIUrl":"https://doi.org/10.1145/3424978.3425009","url":null,"abstract":"For agricultural special species, the labeled procedure of large-scale samples is costly, thus, the bamboo species only has a limited number for supervised learning. The fine-tuning strategy is important for deep neural network by transferring learning methods, which utilize the weight of the deep model of the source domain, and can solve the problem associated with insufficient samples to make the model more stability and robustness. In the manuscript, the novelty of the strategy, for images of bamboo species with low-shot classification, mainly proposed an idea that is the transfer of the convolutional group features of deep convolutional models. The deep models with a novel fine-tuning method and three optimizers that are stochastic gradient descent, Adaptive Moment estimation, and Adadelta respectively, are evaluated by the accuracy and the expected calibration error value for the analysis of deep model generalization. An analysis of the results showed that, based on the proportion of training dataset is only 30%, the innovative strategy for bamboo species classification achieved better performance that has an accuracy of 0.82, and the expected calibration error of 0.16, which were better stability and generalization than those of other fine-tuning strategies. Consequently, the novel fine-tuning strategy proposed in this manuscript transfers the features of deep convolutional groups, improves the accuracy and generalizability of the model, and resolves the problems associated with having insufficient samples of bamboo species for low-shot classification.","PeriodicalId":178822,"journal":{"name":"Proceedings of the 4th International Conference on Computer Science and Application Engineering","volume":"4 10","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2020-10-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133052713","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：