摘要
蟹苗的密度估计在蟹苗养殖中有着重要的意义。但现有的基于图像处理的密度估计算法不能对蟹苗图像进行准确地密度估计,因此提出一种基于图像特征与竞争型神经网络的蟹苗密度估计算法。首先引入背景建模与噪声处理得到前景图像像素数;然后使用阈值区分高密度图像和低密度图像;最后对阈值上下图像分别采用基于全局特征的密度等级分类算法和基于局部特征的线性回归算法。其中蟹苗图像的密度等级由竞争型神经网络进行划分和判断。实验结果表明正确率可达到93.95%。
Crab larvae density estimation is of great significance in breeding crabs. However, existing density estimation algorithms based on image processing cannot accurately estimate the density of crab larvae images at different density levels, we propose a crab density estimation algorithm based on image feature and competitive neural network. First we introduce background modeling with noise processing to get foreground image pixels. Then we use thresholds to distinguish between high density images and low density images. Finally, a regression algorithm based on local features is used to analyze images below the threshold and a classification algorithm based on global features is used to analyze images above the threshold. Density levels of crab larvae image are divided and judged by competitive neural network. Experimental results showed that algorithm can accurately estimate the density of crab larvae images at different density levels.
出处
《计算机应用与软件》
2017年第8期236-240,共5页
Computer Applications and Software
关键词
蟹苗密度估计
阈值判别
线性回归
竞争型神经网络
纹理特征
Crab larvae density estimate Threshold determination Linear regression Competitive neural network Texture feature