基于神经网络的茄子采摘机器人视觉识别方法

针对生长环境中茄子图像背景复杂的特点，提出了一种基于BP神经网络的图像分割方法。通过对茄子果实的分析，选取3×3邻域像素EXG灰度值作为图像特征。选取30幅图像作为训练样本，以人工借助Photoshop软件分割后的图像作为教师信号，采用改进的BP算法对神经网络的权值进行训练。经过120次循环后，获得有效的网络权值，误差为0．001。结果表明，利用BP神经网络能够较好地实现茄子与背景的分离，经过数学形态法结合中值滤波方法的进一步处理后完全能满足采摘机器人的要求。

Target Identification Method of Eggplants Picking Robot Based on Improved BP Neural Network

Aim at the complex background of eggplant image in the growing environment, a image segmentation method based on BP neural network was put forward. The EXG gray values of 3 × 3 neighborhood pixels were obtained as image fealures through by analyzing the eggplant image. 30 eggplant images were taken as training samples and results of manual segmentation images by Photoshop were regarded as teacher signals. The improved BP algorithm was used to train the parameter of the neural network. The effective parameter was achieved after 120 times of training. The result of this experiment showed that the eggplant fruit could be preferably segmented from the background by using BP neural network algorithm and it could totally meet the demands of the picking robots after further processing by way of combining mathematics morphology with median filtering.