儿童手写运动分析方法及其应用研究

发布时间：2018-05-02 22:42

本文选题：手写运动 + 数字手写板　；参考：《中国科学技术大学》2017年博士论文

【摘要】：手写运动由于蕴含丰富的认知、生理以及病理信息而被广泛用于认知特点、精细运动控制机制和相关疾病病理的研究之中。尽管围绕手写运动展开了大量的基础性研究工作,但主要还是集中在疾病群体手写特点分析上,对健康群体的关注较少,而且对手写特征的描述较为局限,同时也缺乏应用层面的手写运动分析探究。因此,本研究分别选取健康儿童和以自闭症为代表的疾病儿童为研究对象,基于新的手写稳定性维度,结合传统特征进行手写特点分析;同时,在此基础上引入机器学习算法,实现手写特征在疾病对象量化分类过程中的有效应用。论文的主要研究内容包括:1.健康儿童汉字书写特点研究了解健康儿童手写特点对儿童书写过程中的教学指导,以及书写困难的量化评估具有重要意义。目前,对于健康儿童的手写研究关注较少,而且,缺乏基于写字任务的中国儿童写字发展特点研究。因此,本文采用汉字书写任务,分析1-5年级儿童手写发展趋势。基于重复性书写活动,提取手写特征的标准差进行稳定性度量;同时,引入动态时间规整算法(Dynamic time warping,DTW)进行书写轨迹及速度特征序列的稳定性评估。研究结果表明,随着年级的变化,儿童以非线性的方式逐渐掌握手写技能,1、2、3年级是手写能力变化快速发展时期。签名能力的提高主要体现在,书写时间与运动幅度的下降,手写压力增大与手写稳定性提高。本文提取的包括特征标准差和DTW距离在内的稳定性特征的变化,验证了其在手写稳定性量化上的有效性。2.自闭症儿童手写特点分析自闭症是儿童书写困难的高发群体,但针对自闭症手写特点的量化研究较少。此外,通过对健康儿童手写稳定性的量化研究,了解了儿童手写稳定性发展特点,但疾病对于手写稳定性的影响仍不明确。本文采用阿拉伯数字书写任务首次分析中国自闭症儿童的手写表现。通过分别提取自闭症儿童和正常对照组儿童书写时的时空、运动学与动力学特征展开手写系统研究。采用重复书写任务提取包括DTW距离在内的稳定性指标。数据分析表明:自闭症儿童手写的异质性主要表现在数字序列任务中,较大的书写尺寸、较高的书写速度与加速度;以及重复书写任务中,较差的稳定性。手写特征在不同任务中的差异性验证了手写任务对手写表现的影响,结合自闭症的大脑神经机制,发现自闭症儿童可能在复杂的手写运动中存在运动障碍。研究结果进一步验证了包括DTW距离在内的稳定性特征在手写评估中的有效性,同时为自闭症有效教学及干预训练提供了参考。3.手写量化分类研究疾病的辅助诊断是手写应用中重点关注的问题,但目前针对实际应用问题开展的手写运动分析研究有限。为了实现手写在自闭症量化分类中的有效应用,本文基于对自闭症儿童手写特点的分析结果,采用机器学习算法进行手写量化分类研究。为了提高自闭症群体的分类准确度,从图像的角度出发,首次尝试提取能够表现自闭症儿童手写局部信息的图像特征;结合传统的手写时空特征、运动学特征,提出了一种基于特征融合的度量学习算法。实证分析不同特征提取方法对分类准确率的影响,同时对比了经典的分类算法与本文的基于多特征的度量学习算法在分类方面的表现。结果表明,使用本文提出的多特征度量学习算法,组合传统特征与图像特征,能够实现对自闭症儿童更好的分类效果。本文提出的算法模型,为手写运动分析在自闭症及其他相关疾病的辅助诊断与综合评估上的有效应用提供了一个新的技术途径。
[Abstract]:Handwritten motion is widely used for cognitive, physiological and pathological information, which is widely used in the study of cognitive characteristics, fine motion control mechanism and related disease pathology. Although a large number of basic research work has been carried out around handwriting movement, it is mainly focused on the analysis of handwriting characteristics of disease groups and on health groups. There are few notes and limited description of handwritten features and lack of application level handwritten motion analysis. Therefore, this study selected healthy children and children with autism as the research object, based on the new handwritten stability dimension, combined with traditional characteristics to carry out handwriting characteristics analysis; at the same time, the basis of this study is the basis of handwriting characteristics analysis. The application of machine learning algorithm is introduced to realize the effective application of handwritten features in the process of quantifying and classifying disease objects. The main contents of this paper are as follows: 1. the characteristics of Chinese character writing in healthy children is of great significance to the teaching guidance in the process of children's writing and the quantitative evaluation of the difficulty of writing. There is less attention to the handwriting study of healthy children, and there is a lack of research on the characteristics of Chinese children's writing development based on the writing task. Therefore, this paper uses the Chinese character writing task to analyze the development trend of the 1-5 grade children's handwriting. Based on the repetitive writing activity, the standard deviation of handwriting features is extracted and the dynamic time is introduced. Dynamic time warping (DTW) is used to evaluate the stability of writing trajectory and speed feature sequence. The results show that, with the change of grade, children gradually master handwriting skills in a nonlinear way. The 1,2,3 grade is the period of rapid development of handwriting ability. The improvement of signature ability is mainly reflected in the writing time and transportation. The decrease of dynamic amplitude, the increase of handwriting pressure and the improvement of handwriting stability. In this paper, the characteristics of stability, including the standard deviation and the DTW distance, are extracted, and the validity of the handwritten stability quantizing is verified by the handwriting characteristics of.2. autistic children. There are few quantitative studies on writing characteristics. In addition, by quantifying the handwritten stability of healthy children, the characteristics of the development of handwritten stability of children are understood, but the influence of disease on handwriting stability is still unclear. In this paper, the handwriting performance of Chinese autistic children was first analyzed by the Arabia digital writing task. The time and space, kinematics and dynamics of children and normal control children were studied by handwriting system. The stability indexes including DTW distance were extracted by repeated writing tasks. Data analysis showed that the handwritten heterogeneity of autistic children was mainly in the digital sequence task, the larger writing size, the higher book. Writing speed and acceleration; and the poor stability in repeated writing tasks. The difference between handwriting features in different tasks verifies the effect of handwritten tasks on handwriting performance, combined with the brain neural mechanism of autism, and finds that autistic children may have dyskinesia in complex handwriting exercises. The results are further verified. The effectiveness of the stability features including the DTW distance in the handwriting evaluation, and the effective teaching and intervention training for autism provides a reference.3. handwritten quantization classification study of the auxiliary diagnosis of disease is a key concern in handwriting applications, but the current handwritten motion analysis for practical applications is limited. The effective application of handwriting in the quantitative classification of autism. Based on the analysis of the handwritten characteristics of autistic children, this paper uses machine learning algorithm to carry out handwritten quantization classification. In order to improve the classification accuracy of the autistic group, from the angle of the image, the first attempt to extract the handwritten local letter for children with autism is first tried. In combination with traditional handwritten space-time features and kinematic features, a measurement learning algorithm based on feature fusion is proposed. The effect of different feature extraction methods on classification accuracy is analyzed, and the classification performance of the classical classification algorithm and the multi feature based measurement learning algorithm in this paper is compared. The results show that the use of the proposed multi feature metric learning algorithm, combining the traditional features and image features, can achieve a better classification effect for autistic children. The proposed algorithm model provides a new method for the effective application of handwritten motion analysis in the auxiliary diagnosis and comprehensive evaluation of autism and other related diseases. Technical approach.

【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R749.94;TP181

【相似文献】