多对象特征提取和优化神经网络的玉米种子品种识别

为了实现机器视觉代替人的视觉，对玉米种子品种进行实时、客观、准确和无损伤识别，研制了玉米品种识别硬件系统和软件系统。针对玉米种子及种子图像的特点，对玉米种子品种识别技术与算法进行了深入地研究和探索，提出了一种基于多对象有效特征提取和主成分分析优化神经网络的玉米种子品种识别方法，提取了玉米种子的几何特征和颜色特征参数，优化了基于机器视觉的玉米种子图像处理策略和品种识别算法，提高了玉米品种识别的速度和准确率。对农大108、鲁单981、郑单958、五岳18共4个品种玉米种子进行了品种识别试验，每粒种子识别的平均耗时为 0.127 s，综合识别率达到97%以上。研究表明，基于机器视觉的玉米种子品种识别与检测方法是可行的，该方法可提高玉米种子品种识别效率和正确率。

Maize seeds varieties identification based on multi-object feature extraction and optimized neural network

In order to apply machine vision technology replacing human vision to identify maize seed varieties in a real-time, objective, accurate and non-invasive procedure, the hardware and software systems to identify the seeds of maize need to be developed. For maize seed and characteristics of the seed images, the identification technology of maize seed varieties and algorithms has studied and explored in depth. A multi-object features extraction and the optimized neural network using PCA identification method adapting to maize seeds varieties identification was proposed. Geometric features and color features parameters of maize seeds were extracted. Maize seeds image processing strategies and varieties identification algorithms, which was based on the machine vision, was optimized. The precision and speed of maize seeds varieties identification was improved. Through maize seeds varieties identification test on four species including Nongda 108, Ludan 981, Zhengdan 958 and Wuyue 18, average identification time-consuming of each seed was 0.127 s, and integrated identification accuracy was more than 97%. Research shows that identification and detection of maize seeds varieties based on machine vision is feasible, and this method can improve the efficiency and correct identification rate of maize seed varieties.