基于SVM算法的目标分类筛选方法研究

Research on strawberry classification and screening method based on SVM algorithm

针对现有草莓分类工作量大、准确率低的问题,文章结合SVM算法提出一种多特征草莓分类模型。草莓等级划分参照日本宫崎县草莓上市标准,根据草莓周长、面积、形状、颜色、饱和度、成熟度6类特征划分为特等品、一等品、二等品、不合格4类,草莓的形状特征借助傅里叶描绘子定义,草莓的饱和度与成熟度利用HSV颜色模型定义。该研究首先收集拍摄的草莓图片,对图片进行颜色空间转换、图像去噪、目标分割等预处理,然后进行各类特征的提取,最后将提取出的特征经过整理与算法训练,通过SVM算法对草莓进行分类。经测试,该筛选方法的综合正确率为84.34%,实验综合运行时间为58 ms,表明可以通过SVM算法达到识别草莓等级的目的,且具备实时性。

Aiming at the problems that the existing strawberry classification is affected by human factors, the workload is huge and the accuracy is low, a new multi-class feature model combined with SVM algorithm is proposed to classify strawberries. The classification of strawberries refers to the listing standards of strawberries in Miyazaki Prefecture,Japan. According to the six characteristics of strawberry perimeter, area, shape, color, saturation and maturity, it is divided into four categories: special, first-class, second-class, and unqualified. The shape characteristics of strawberries are defined by Fourier descriptors, and the saturation and ripeness of strawberries are defined by the HSV color model. After the strawberry pictures are collected, the strawberry images are subjected to preprocessing such as color space conversion, image denoising, and target segmentation, and then various features are extracted. The extracted features are sorted and trained by the algorithm, and the strawberries are classified by the SVM algorithm.After testing, the comprehensive correct rate is 84. 34%, indicating that the purpose of identifying the strawberry grade can be achieved through the SVM algorithm, and the comprehensive running time of the test is 58ms, indicating that it has real-time performance.