基于软件工程与叠层深度学习的工件文本识别算法

Workpiece text recognition algorithm based on software engineeringand stack in-depth learning

目的解决当前图像特征训练不充分,系统软件不规范,导致文本识别不准确的问题。方法分别从系统软件开发和识别算法验证的角度出发,提出基于软件工程与叠层深度学习的工件文本识别算法。首先,根据系统功能需求,进行软件模块化分析,设计出集算法计算、硬件控制、逻辑通信和数据存储于一体的系统架构。然后,基于自适应阈值分割与图像校正,对工件文本图像进行预处理,得到准确的包含文本目标的二值图像区域。最后,利用叠层DAE构成L层深度网络,计算权值矩阵,达到对文本图像轮廓特征深度训练学习的目的。结果利用所提算法获得了复杂干扰条件下的文本识别结果。结论实验测试结果显示:与当前文本识别技术相比,本文算法拥有更高的准确性与现场实用性。

Purposes-To solve the problem of inaccurate text recognition which is caused by insufficient training of the current image feature and nonstandard system software. Methods-The text recognition algorithm based on software engineering and stack in-depth learning is proposed from the point of view of system software development and recognition algorithm verification. First of all, the software module is analyzed on the basis of the functional requirements of the system, and the system architecture is designed, which is integrated with the algorithm calculation, hardware control, logic communication and data storage. Secondly, the text image of the workpiece is pretreated on the basis of the adaptive threshold segmentation and image correction, and the binary image region containing the text target is obtained accurately. Finally, the laminated DAE is used to form the deep network of Layer L, and the weight matrix is calculated to achieve the purpose of the deep training of the text im- age contour feature. Results-The proposed algorithm is used to obtain the text recognition results under complex interference conditions. Conclusions-The experimental results show that the system has higher accuracy and practicability in comparison with the current text recognition system.