一种采用支持向量机和凸包拟合的茄子识别方法

传统的茄子图像识别研究大多数针对单果、无遮挡、自然环境较简单的情况，而解决复杂自然环境下多果、遮挡的茄子识别问题，已经成为茄子采摘机器人急需解决的问题。对于无遮挡的情况，采用支持向量机进行分割，并且应用开运算去除细小连接。为了去除大面积噪声，采用面积法和外接矩形法。针对背景与茄子相似的情况，采用直方图匹配的方法进行分割识别。对于被遮挡的茄子应用凸包拟合的方法进行识别。最后，与其他算法在单果、多果、遮挡、背景复杂、表面反光、总识别率这6种情况下进行比较，结果表明，该算法的识别率较高。

A method of eggplant recognition using support vector machine and convex hull fitting technology

Most of the traditional researches of image of eggplant recognition solve the problem of single fruit, unobstructed and relatively simple natural environment. The eggplant recognition, which includes multiple fruit, occluded fruit and the complex natural environments has become the urgent problem of the eggplant harvesting robot. For the case of unobstructed, support vector machine is applied to segment. And then the open operation is applied to remove the small connection. The area method and the method of bounding rectangle are used to remove the noise of a large area. The similar situation for the background and eggplant, the method of histogram matching is used to segment and recognize it. The method of convex hull fitting is used to segment and recognize the situation of occluded eggplants. Finally, the experimental results show that the proposed algorithm has a better recognition rate compared with other algorithms in six conditions of single eggplant, multiple eggplant, occluded eggplant, complex background, surface reflectivity, total recognition rate.