基于BP神经网络的大豆种子外观品质质量检测技术研究

Research on Soybean Seed Appearance Quality Testing Technique Based on BP Neural Network

为了提升传统大豆种子质量检测水平,保证检测结果的准确性,采用图像处理方法,建立精确的BP神经网络诊断模型,分析了BP神经网络结构和学习训练过程,对特征参数进行了主成分析法降维处理,采用BP神经网络的训练和测试,对单类大豆种子识别率很高。针对实际生产,通过比较不同共轭梯度算法,采用Levenberg-Marquardt算法训练网络,误差更小,通过反复训练,使其具备准确识别并分级大豆种子质量的能力。

In order to improve traditional quality testing level of soy seeds,and guarantee the accuracy of testing results,the picture processing method is adopted,and accurate BP neural network diagnosis model is established.The research analyzes the structure of BP neural network,and learning and training process.Characteristic parameters are disposed with principal component analysis method for dimensionality reduction,and training and test of BP neural network are adopted.So the recognition rate of a single class of soybean seeds is high.According to practical production,through the comparison of different conjugate gradient algorithms,Levenberg-Marquardt arithmetic training network has less error.Through repeat training,it can be qualified with the capacity of accurate recognition and classification of soybean seed quality.