作业是小学生学习活动的重要组成部分，对于学业成绩、习惯养成等意义重大，而作业专注与分心情况也是研究者和一线教师、家长关心的话题。而以往的专注度测量方法或不适用于小学生作业场景使用，或需要侵入性较强且昂贵的仪器，因此，本研究旨在通过摄像头便捷地采集视频数据用于分析小学生作业专注度，并基于此设计识别模型和干预系统，从而提升小学生作业专注水平，帮助其养成良好的学习习惯，辅助家长和教师的教育过程。本研究主要包括以下四部分：

通过文献梳理和实际观察，对小学生作业分心行为进行分类，并制定了小学生作业低专注度行为观察编码表，专家背对背编码在一致性检验中Kappa系数为0.80，证明该编码表效果较好。

根据编码表，在实验采集的80名小学生的作业视频数据集上，使用基于ImageNet的ResNet50神经网络模型预训练方法，构建了小学生作业低专注识别模型，能以秒为单位实时输出小学生作业专注情况“专注”或“非专注”，并给出相应的置信度（0-100%），准确率能达到90.20%，识别效果较好。

通过调研分析目前已有专注度干预工具的相关功能，整理出低专注的机器干预策略；通过对小学教师和小学生家长进行访谈，梳理出小学生作业分心常见原因、人工干预方法、以及教师家长期望的机器干预功能。根据以上调研结果设计了六类人机结合的小学生作业低专注干预方法，包括：提醒、时间规划、激励机制、数据记录、辅导、其他等。

基于调研和访谈结果，提出了小学生作业低专注干预系统设计原则：功能简洁明了，避免额外干扰；监测反馈适度，尊重学生隐私；注重人机交互，精准解决问题。并设计了小学生作业低专注干预系统原型，系统有学生端、家长端、教师端等三种模式；依据作业过程的不同步骤，系统功能包括学习计划、学习过程、学习评价三大模块，涵盖了小学生作业过程中11个常见场景。

Homework is an important part of primary school students' learning activities, which is of great significance to academic performance and habit formation, and homework concentration and distraction are also topics of concern to researchers, front-line teachers and parents. However, the previous measurement methods of concentration may not be suitable for primary school students' homework scenes, or require invasive and expensive instruments. Therefore, this study aims to collect video data conveniently through the camera to analyze primary school students' homework concentration, and design an identification model and intervention system based on this, so as to improve primary school students' homework concentration level, help them develop good study habits and assist parents and teachers in the educational process. This study mainly includes the following four parts:

Based on literature review and practical observation, this paper classifies primary school students' distracted behaviors in homework, and formulates a code table for observing primary school students' behaviors with low concentration in homework. The Kappa coefficient of expert back-to-back coding is 0.80 in consistency test, which proves that the code table is effective.

According to the coding table, based on the experimental video data set of 80 primary school students, using the pre-training method of ResNet50 neural network model based on ImageNet, a model for identifying the low concentration of primary school students' homework is constructed, which can output the "concentration" or "non-concentration" of primary school students' homework in seconds in real time and give the corresponding confidence (0-100%), with an accuracy rate of 90.20% and a good recognition effect.

Through the investigation and analysis of the related functions of the existing concentration intervention tools, the low concentration machine intervention strategy is sorted out; Through interviews with primary school teachers and parents of primary school students, the common causes of primary school students' homework distraction, manual intervention methods and the machine intervention function expected by teachers and parents are sorted out. According to the results of interviews and tool research, six kinds of intervention methods for low concentration of primary school students' homework are designed, including: reminder, time planning, incentive mechanism, data recording, counseling and others.

Based on the results of investigation and interview, this paper puts forward the design principles of low concentration intervention system for primary school students' homework: the function is concise and clear, avoiding extra interference; Appropriate monitoring feedback, respect for students' privacy; Pay attention to human-computer interaction and solve problems accurately. And designed the prototype of primary school students' homework low concentration intervention system, which has three modes: students' end, parents' end and teachers' end. According to the different steps of the homework process, the system functions include three modules: learning plan, learning process and learning evaluation, covering 11 common scenes in the homework process of primary school students.