生物形态学图像分析系统

BIOMORPHOLOGICAL IMAGE ANALYSIS SYSTEM

周睿洋

11749195

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

生物形态学是研究动物、植物和微生物组成部分的外形和结构的科学。生物形态学研究的重要基础是快速高效地提取生物生长状态地形态学信息，并进行量化分析。由于生物生长速度慢、时间跨度较长。这个分析过程通过传统人工测量方法效率低、误差大，难以长期进行。传统显微镜镜检中检测人员通过肉眼在显微镜下直接进行观察，效率低，检测准确度随着工作时间增加出现下降，需要对样本进行多次重复的检验，耗费大量人力物力。传统镜检方法已经难以适用于生物形态学方面的检测。当前正处在计算机软硬件高速发展的阶段，计算机硬件更新迅速，硬件运算速度大幅度提高。通过GPU的高速运算，为计算机图像快速处理提供了硬件保证。同时，图像处理技术经过多年发展，已经日趋成熟。而基于深度学习的图像神经网络的大量研究成果，如RCNN、yolo、SSD等神经网络，将图像识别准确度、可靠性大幅度地提高，机器视觉可以应用于自动化生产生活领域，基于机器视觉的生物形态学快速准确分析识别成为可能。本文的研究工作主要应用计算机视觉对生物形态进行测量分析识别，通过显微镜拍摄植物和微生物图片，将传统图像方法与深度学习结合，对显微图片进行识别、测量，并通过vc++对形态学处理系统进行实现。研究工作主要分为两方面，对放置在培养基上的植物幼苗根部生长过程进行相同时间间隔的连续拍摄，通过传统图像处理方法，提取根部各组成部分的生长信息并绘制生长状态曲线，作为根须生长的量化指标；以及结合根须生长特征，使用分层拍摄的方式对植物根部进行拍摄，通过传统方法与yolo神经网络结合，将根须拍摄的图片通过传统方法进行提取、分割、增强后传入yolo 神经网络进行识别及计数，并在分析后绘制图表及处理结果报表。论文首先在绪论中介绍本课题的背景，随后介绍处理系统所使用的传统方法如图像平滑、阈值分割、腐蚀膨胀、边缘检测及细化算法和深度学习算法yolo神经网络，以及分析软件的具体设计和系统中所用到的并行化等方法。最后对研究工作进行总结，并提出未来展望。本文对植物根须形态测量以及微生物识别检测提供了一种新的方法，并为后续的生物形态学研究及工业生产提供了快速检测的方法，在准确率和速度方面有较高的应用价值，可以为后续这方面研究提供参考和借鉴。

Biomorphology is the science of studying the shape and structure of animal, plant and microbial components. An important basis for biomorphological research is the rapid and efficient extraction of morphological information from biological growth states and quantitative analysis as required. Due to the slow growth of organism and the long time span, this analysis process is difficult and long-term through traditional manual measurement methods with low efficiency and lots of errors. In tradition microscopy, the tester directly observes under the microscope through the eye. This method is inefficient, and the detection accuracy decreases with the increase of working time. It is necessary to repeatedly test the sample and consume a lot of manpower and material resources. It can be seen that traditional microscopy methods have been difficult to apply to biomorphological testing. At present, the computer hardware and software are developing at a high speed, the computer hardware is updated rapidly, and the hardware operation speed is greatly improved. Through the high-speed operation of the GPU, hardware guarantee is provided for the computer to process images at high speed. At the same time, image processing technology has matured over the years. The research gain of image neural network based on deep learning, such as RCNN, yolo, SSD and other neural networks, greatly improve the accuracy and reliability of image recognition, so that computer vision can be applied to the field of automated production and life, which makes the computer vision of biomorphology possible to analyze and identify quickly and accurately. The research work in this paper mainly uses computer vision to measure and analyze biological morphology, photograph plants and microbe through microscope, combine traditional image method with deep learning, identify and measure microscopic images, and implement the algorithm through vc++. The research work is mainly divided into two parts, the roots of the plant seedlings placed on the medium were continuously photographed at the same time interval, and the growth information of each component of the roots was extracted and the growth state curve was drawn by the traditional image processing method as a quantitative index of root growth; and combining root growth characteristics, photographing plant roots using layered photography. The image processing method combined with traditional image processing and deep learning. The images taken by microbial scanning are extracted, segmented and enhanced by traditional methods and then transmitted to the yolo neural network to identify and count. Chart and report generates after analyzing. The paper first introduces the background of the subject in the introduction, and then introduces the traditional methods used in the processing system such as image smoothing, threshold segmentation, dilation and erosion, edge detection, refinement algorithms and deep learning yolo network. Then introduce the specific design of the software and parallelization methods used in the system. Finally, the research work is summarized and the future prospects are proposed. This paper provides a new method for determination of root shape and microbial morphology of plants, and provide a rapid detection method for subsequent biomorphological research and industrial production, which has high application value in accuracy and speed. It can provide reference for subsequent research in this area.

机器视觉 生物形态学 图像处理 深度学习 神经网络

vc++ computer vision biomorphology image processing deep learning neural network vc++

中文

联合培养

南方科技大学

周睿洋. 生物形态学图像分析系统[D]. 深圳. 哈尔滨工业大学,2019.