摘  要

近些年来，基于人工智能技术及其创新应用的教学是计算机类人才培养的重要环节。玉米果穗智能筛分是人工智能在农业领域的重要应用，可作为典型案例融入到人工智能教学领域中。本文将人工智能技术与农业育种应用需求结合，以“原理认知-算法实践-实际案例应用”层层递进的教学模式，开发了一套玉米果穗高通量智能筛分虚拟仿真实验系统，主要工作如下：

（1）利用Unity3D引擎的UGUI插件设计良好的用户交互界面，通过三维交互方式展现复杂的人工神经网络知识点及模型结构。

（2）基于人工神经网络的智能分析技术，搭建智能识别模型及算法分析仿真界面，通过WebSocket协议调用百度Paddle2.0框架下多种深度神经网络模型（AlexNet、VGG、ResNet和MobileNet等）进行仿真训练，用户通过不同模型及参数的组合应用，提升调参优化模型能力。

（3）利用三维建模软件3ds Max建立玉米果穗筛分控制装置及三维场景相关模型。结合虚拟仿真技术和人工智能技术，多方位还原玉米育种筛分流水线典型工业现场功能模块。利用人机交互技术，通过鼠标键盘操作，实现流水线的虚拟场景和硬件模块细节的浏览。同时引入基于人工智能玉米果穗异常识别技术，采集真实的玉米果穗图像进行分类识别，提高典型智能控制系统应用场景的训练效果。

（4）此外，整个系统利用MySQL进行数据库设计，为系统提供数据支持。最后从功能和性能方面对系统进行了测试工作，验证系统是否达到系统目标。

由于本系统以实际农业育种的科研问题为背景，因此可以更有效地培养学生解决人工智能领域复杂工程的能力，后续可以在人工智能教学，特别是农业、自控相关的人工智能教学中起到重要作用。

关键词：玉米果穗筛分，虚拟仿真，人工智能，深度学习

Abstract

In recent years, teaching based on artificial intelligence technology and its innovative application is an important link in the training of computer talents. Intelligent sieving of corn ears is an important application of artificial intelligence in agriculture and can be used as a typical teaching case.  In this paper, combining artificial intelligence technology with application requirements of agricultural breeding, a virtual simulation experimental system for high-throughput intelligent screening of corn ears was developed based on the progressive teaching mode of "principle cognition - algorithm practice - actual case application". The main work is as follows: (1) UGUI plug-in was used to design a good user interaction interface, and complex artificial neural network knowledge points and model structure were displayed through 3D interaction. (2) Based on the intelligent analysis technology of artificial neural network, intelligent recognition models and a simulation interface of algorithm analysis were built, and WebSocket was uesd to call a variety of deep neural network models (AlexNet, VGG, ResNet and MobileNet, etc.) under the Paddle2.0 framework for simulation training. Through the combined application of different models and parameters, users can improve the ability of model parameter tuning. (3) 3ds Max was used to establish the corn ears screening control device and related models. Combining virtual simulation with artificial intelligence technology, multi-directional restoration of typical industrial field functional modules of screening assembly line. Using human-computer interaction, through the mouse and keyboard operation, the virtual scene and the details of module were realized.  At the same time, the artificial intelligence-based corn ears abnormal recognition technology was introduced to collect real images for classification, improving the training effect of typical intelligent control system application scenarios. (4) In addition, MySQL was used for database design to provide data support. Finally, the system was tested from the aspects of function and performance to verify whether the system achieved the goal.

As the system was based on the scientific research problems of actual agricultural breeding, it can more effectively cultivate students' ability to solve complex engineering in artificial intelligence, and it can play an important role in artificial intelligence teaching, especially in the artificial intelligence teaching related to agriculture and self-control.

Key words: Corn ears screening, Virtual simulation, Artificial intelligence, Deep learning