小鼠行为学自动化分析系统

AUTOMATIC ANALYSIS SYSTEMS FOR MOUSE BEHAVIOR

邵开

11749194

硕士

计算机技术

陈放怡

2019-05-21

2019-06-28

哈尔滨工业大学

深圳

在大部分的哺乳动物中基本上都有自发的活动和社会行为等特征。由于小鼠的行为和活动能够定量的反映出小鼠对于外界环境和药物的反应。所以自发性活动和社会行为的统计信息，对于生物医学等领域有着重要的意义。早期的动物行为学的研究主要是依赖人工的标记，这样会导致实验的大部分时间在进行重复的标记工作，并且人工标记具有一定的主观性和偏差，对于参数定量化也存在着困难。近几年随着计算机视觉和图形学技术的发展，在小鼠监控和活动行为的自动检测识别上面有了很大的进步。可以通过图像处理的方法，自动的处理视频中小鼠的行为和活动。OPENCV图像库的发展让识别任务变得更加便捷。由于拍摄环境的变化时刻在发生着改变，所以特征提取的工作变得很繁琐，对于一些复杂的环境进行物体特征提取的结果很不稳定。所以OPENCV主要来做一些图像的预处理部分。近两年随着计算机处理速度的发展和深度学习BP算法的发现，引发了神经网络的第二次浪潮。神经网络模型经过10几年的不断完善和改进，在图像识别、自然语言处理、视频处理等方便都有了突破性的进展。本文主要采用了传统的图像预处理方法和YOLO模型相结合的方法，对小鼠的运动和行为的视频信息进行自动化处理。通过小鼠转圈实验来对小鼠的前庭损伤进行定量的分析。在黑暗环境和正常环境下小鼠活动之间的差异性来评判小鼠对不同环境的反映情况。患自闭症小鼠和正常小鼠的活动范围和独处时间来判断自闭症的患病程度。通过以上综合的方法可以对正常小鼠和患病小鼠的自发行为和群体行为进行定量的分析过程，从而更加准确、客观的为实验结果提供数据证明。

In most mammals there is basically spontaneous activity and social behavior and other characteristics. Since the behavior and activity of mice can reflect quantitatively the response of mice to the external environment and drugs. Therefore, the statistical information of spontaneous activities and social behaviors is of great significance to the fields of biomedicine. Early studies on animal behavior mainly rely on artificial markers, which will lead to repeated labeling for most of the time of the experiment. Moreover, artificial markers are subjective and biased to some extent, and it is difficult to quantify the parameters. In recent years, with the development of computer vision and graphics technology, great progress has been made in the automatic detection and recognition of mouse monitoring and activity behavior. The behavior and activity of video mice can be automatically processed by image processing. The development of OPENCV image library makes the recognition task more convenient. Because the filming environment is changing all the time, the work of feature extraction becomes very complicated, and the result of object feature extraction in some complex environments is very unstable. So OPENCV mainly to do some image preprocessing part. In the past two years, with the development of computer processing speed and the discovery of deep learning BP algorithm, the second wave of neural networks has been triggered. After more than 10 years of continuous improvement and improvement, neural network model has made a breakthrough in image recognition, natural language processing, video processing and other convenience. In this paper, the traditional image preprocessing method and YOLO model were combined to automate the processing of video information of the mice's movement and behavior. VOR and mouse circle experiments were used to quantitatively analyze vestibular injuries in mice. The difference between mice in the dark environment and the normal environment was used to evaluate the mice's response to different environments. The extent of autism was determined by the range of motion and time spent alone in both the autistic and normal mice.Through the above comprehensive methods, the spontaneous and group behaviors of normal mice and sick mice can be quantitatively analyzed, so as to provide more accurate and objective data evidence for the experimental results.

小鼠行为 深度学习 opencv 图像处理

mouse behavior deep learning opencv image process

中文

联合培养

南方科技大学

邵开. 小鼠行为学自动化分析系统[D]. 深圳. 哈尔滨工业大学,2019.