论文部分内容阅读
在农业领域,实现自动、准确、稳健的种子分类识别算法是具有重要经济意义的.由于杂草种子的类别很多,大小、形状、纹理等特征变化多样,即使一个类别的种子,不同的特征也会在数量上有所差异.另外,由于种子常常会因潮湿、病菌等因素,产生霉块或病斑即是连续遮挡或噪声的问题,而之前的算法不适合解决此问题.文中通过求解一个复合的欠定线性方程组优化问题来解决杂草种子的连续遮挡问题.文中利用压缩感知理论在机器学习领域的运用,用主成分分析、下抽样、随机取样等方法对种子图像降维,然后就把杂草种子分类问题归结为一个求解待测样本对于整体训练样本的稀疏表示问题.问题的求解通过e~1范式最小化完成.实验结果表明,利用稀疏表示算法进行分类,可以达到很好的识别效果,对于87类的杂草种子,最好的识别率是90.80%.
In the field of agriculture, it is of great economic significance to realize an automatic, accurate and robust seed classification recognition algorithm.Due to the large variety of weed seeds, features such as size, shape and texture vary so much that even one type of seed, different characteristics In addition, because the seeds often due to damp, germs and other factors, mold or disease spots that are continuous occlusion or noise problems, and the previous algorithm is not suitable to solve this problem.In this paper, by solving a Compound underdetermined linear equations optimization problem to solve the continuous occlusion problem of weed seeds.Using the application of compressed sensing theory in the field of machine learning, principal component analysis, downsampling, random sampling and other methods to reduce the dimension of the seed image, and then The weed seed classification problem is reduced to a solution to the sparse representation of the sample to be tested for the whole training sample.The problem is solved by minimizing the e ~ 1 paradigm.The experimental results show that the classification using sparse representation algorithm can achieve very good The recognition rate of the 87 types of weed seeds, the best recognition rate is 90.80%.