TRIZ物场模型和深度学习在蛋黄酥包装图案上的分类研究

TRIZ Field Model and Deep Learning in Classification of Egg Yolk Crisp Packaging Patterns

运用TRIZ物场模型分析目前人工分拣不同口味盒装蛋黄酥的方法,在建立了物场模型后,采用驱动装置和图像采集装置及深度学习算法来替代该物场模型中的人手和人眼,设计了自动分拣不同口味盒装蛋黄酥系统。用改进的Visual Geometry Group(VGG)卷积人工神经网络对盒装蛋黄酥的图案进行分类。由于盒装蛋黄酥图案所在识别环境背景固定、识别角度固定,造成其训练数据集小,因此采用数据增强的方法对数据集进行扩展。在使用1845张蛋黄酥图片进行训练,经过28次迭代后,改进的VGG网络成功收敛。经测试,615张测试样本的分类率都达到100%的正确率。

In this research,TRIZ field model was used to analyze the current manual sorting methods of different flavor boxed crisps.After the establishment of the material field model,the driving device,image acquisition device and deep learning algorithm are used to replace the human hands and eyes in the object field model,an automatic sorting system for different flavors of crispy egg yolk was designed.For the core function visual field of the system,the improved VGG convolution artificial neural network is used to classify the pattern of the packed egg yolk crisp.Because of the fixed background and fixed recognition angle,the training data set is small,so the method of image augmentation is used to expand the data set.After 28 iterations,the improved VGG network converges successfully after using 1845 egg yolk crisp images.The classification rate of 615 test samples is 100%.