王海英1，曹 晶2，谢 骏1，王广军1，胡朝莹1
（1中国水产科学研究院珠江水产研究所，广州 510380；2 广东技术师范学院自动化学院，广州 510630）

摘要：为了将水产养殖水色判别传统技术经验转化为可以量化的数字技术，采用基于L-M神经网络优化算法和计算机图像处理技术的方法，建立了一个水色判别的水产养殖专家系统。通过实例预测，该系统判别误差率＜1%。该系统训练后的神经网络模型，能实现对养殖池塘水质的预测。系统的开发和使用对实现水产健康养殖、智能控制和计算机管理具有一定实用价值．

关键词：水色图像；图像特征值；L-M神经网络优化算法；水质预测

The initially establish of pond water color discrimination system based on the Optimized Algorithm of L-M Neural Network
WANG Hai-ying1, CAO Jing2, XIE Jun1, WANG Guang-jun1, HU Zhao-ying1
( 1 Pearl River Fisheries Research Institute, Chinese Academy of Fishery Sciences, Guangzhou 510380, China;
2 School of Automation, Guangdong Polytechnic Normal University, Guangzhou 510630, China )

Abstract: In order to distinguish the traditional water color of aquaculture technology experience into quantifiable digital technology. An aquaculture expert system for identification of water color was established based on L-M algorithm for neural network optimization and image processing technology. The system conducts water’s color image preprocessing, to extract effective values of image feature, to train model of neural network to achieve the forecast of water quality. The results show that the system identification error rate is less than 1%. The trained neural network model of this system, network model after training of can be able to predict water quality of the breeding ponds. System development and use to the realization of aquatic healthy breeding, intelligent control and computer management have certain practical value.

Key words: water’s color image; image feature values; L-M algorithm for neural network optimization; water quality forecast

基金项目：广东省科技计划项目（2009B020315008）；国家“863”高新技术研发项目（2007AA10Z239）；现代农业产业技术体系项目（NYCYTX-49-13）
作者简介：王海英（1981-），女，硕士，研究方向：水产养殖技术。E-mail: jolly8268@hotmail.com
通讯作者：谢骏（1965—），男，研究员，博士，主要从事水产养殖研究。E-mail: xj007@tom.com

(来源：《渔业现代化》2010.37(5)：19-21)