基于人工智能高碳盘条钢索氏体识别探讨

高碳盘条钢索氏体含量是评价其性能重要指标之一，现行检测方法存在识别准确性差、检测结果容易受检验员影响等缺点。通过索氏体制样标准化、素材收集、素材定值、素材标记等建立供神经网络学习的素材库，利用计算机人工智能和深度神经网络技术初始化识别模型，采用未标记素材对初始化模型进行测试、互动优化，最终验证识别模型的准确率高、检测速度快，识别模型是可行的。人工智能索氏体识别的成功实施，也为晶粒度定级、脱碳层识别、非金属夹杂物识别、带状组织定级等其他金相检测项目的智能识别作出了有益的实践。

Recognition discussion about sorbite in high carbon wire rod steel based on artificial intelligence

The sorbite content of high carbon wire rod steel is one of important indexes for evaluating its performance. The current test methods have some disadvantages, for example, the recognition accuracy is poor, and the detection results are easily affected by the inspectors. A material library for neural network learning was established through sample preparation standardization, material collection, material setting, material labeling and so on. The computer artificial intelligence and deep neural network technology were used to initialize the recognition model. The unlabeled material was used to test and interactively optimize the initialization model. Finally, it proved that the recognition model had high accuracy and fast detection speed. In other words, the recognition model was feasible. The successful implementation of sorbite recognition based on artificial intelligence also provided useful practice for the intelligent recognition of other metallographic testing items such as grain size classification, decarburization layer identification, non metallic inclusion identification, and band structure classification.