基于T-S模型的模糊神经网络在植物病害图像分割中的应用

Fuzzy neural network based on T-S model and its application on image segmentation of plant diseases

针对植物病害图像的病斑区域边缘像素存在模糊性和不确定性,利用T-S模型的模糊规则后件是输入语言变量的函数特性,提出线性清晰化的自适应五层模糊神经网络模型作为植物病害图像模式分类的决策系统,并利用量子遗传算法对模型系统的可调整参数的初始值进行全局优化。试验结果表明:该模型对马铃薯早疫病的彩色图像的有效病班区域分割精确达到100%,学习算法速度快、收敛稳定、鲁棒性较好,避免了传统梯度下降学习算法的局部最小值,并且简单易于实现。

Aiming at the fuzziness and uncertainty of the edge of the lesion region pixels,this paper proposed a decision system of image segmentation of plant diseases with linear and clear self-adaptive five-layer fuzzy neural network model,and optimized the initial values of adjustable parameters by quantum genetic algorithm,which based on the function characteristic that pieces of fuzzy rules of T-S model is the input language variable.The experimental result showed that T-S model had many advantages including accuracy,convergence,stability,robustness,and easy to implement when implied in color image segmentation of potato early blight,which overcame local minimum of traditional gradient descent method.