摘要： 在鱼类育种、渔业资源调查和水产养殖的过程中，均需要对鱼体的体长、体厚、体高等性状参数进行测量。传统测量方法主要采用人工测量方式，劳动强度大、效率低、测量精度低。利用机器视觉技术进行鱼类性状测量可以有效提高测量效率和精度，易于实现自动化。本研究基于机器视觉技术，设计了一个一体化性状测量平台，将鱼类图像采集与分析、体质量与可量性状测量、PIT扫码集成到一个平台，可实现对鲤、鲫鱼性状参数的精确测量。对鱼体2D图像进行性状测量时，需要进行像素校准，由于每条鱼体厚不同，传统的固定像素校准平面在鱼体厚较大时，难以实现精确测量。本研究设计了一种像素校准方法，即在采集鱼体图像时，通过距离传感器测定鱼体体厚，根据体厚大小，确定每条鱼轮廓面的像素校准参数，减小测量误差，提高鱼体可量性状测量的准确性。以鲤、鲫鱼人工测量结果作为对照，比较该系统的测量误差，结果显示，随着鱼体体厚的增加，体长、体高、头长、尾柄长、尾柄高、体厚等参数的相对测量误差没有显著增大。体高和体长相对误差最大值为1.05%、-1.39%；头长、尾柄高、尾柄长和体厚等相对误差最大值分别为1.73%、-2.79%、-2.88%和-2.1%，绝对误差值均小于1 mm。体质量相对误差最大值为-0.36%。该系统满足鲤、鲫鱼性状测量要求。  

关键词: 机器视觉, 鱼类性状, 性状测量, 表型信息, 鲤, 鲫

Abstract: In the process of fish breeding, fishery resources survey, and aquaculture, the length, thickness, and height of the fish body are measured. The traditional measurement method mainly adopts artificial measurement, which has the disadvantages of high labor intensity, low efficiency, and low measurement accuracy. Using machine vision technology to measure fish traits can effectively improve measurement efficiency and accuracy, and is easy to be automated. Based on machine vision technology, we designed an integrated trait measurement platform that  integrated fish image collection and analysis, body weight and weight available trait measurement, and PIT scanning code, which can realize the accurate measurement of trait parameters of carp and crucian carp. Pixel calibration is required for character measurement of 2D fish images. Due to the different body thicknesses of each fish, it is difficult to achieve accurate measurement with a traditional fixed pixel calibration plane when the body thickness is large. To solve this problem, this study designed a pixel calibration method. During the collection of fish body images, the thickness of the fish body was measured by a distance sensor, and pixel calibration parameters of each fish contour surface were determined according to the thickness, so as to reduce the measurement error and improve the accuracy of the measurement of measurable traits of fish. The measurement errors of the system were compared with the manual measurement results. The results showed that the relative measurement errors of body length, body height, head length, caudal stem length, caudal stem height, and body thickness have no significantly change with the increase in body thickness. The maximum relative errors of body height and body length were 1.05% and -1.39% respectively. The maximum relative errors of head length, caudal peduncle height, caudal peduncle length, and body thickness were 1.73%, -2.79%, -2.88%, and -2.1% respectively, and the absolute errors were all less than 1mm. The maximum relative error of body weight was -0.36%. This indicates that the system fits the requirements of fish characters measurement.

Key words: machine vision, fish, characters, measurement, phenotype data, Cyprinus carpio, Carassius auratus