Abstract A Computational Vision System (CVS) based on a Convolutional Neural Network (CNN), operating with the module Inception v3 and reduced number of convolutional layers (Short-CNN), was proposed for the classification of browning degree of bread crust during baking. The training (70%), validation (15%), and testing (15%) of the CNN was performed using a dataset composed of 374 bread crust image fragments (600 × 600 pixels) over seven baking periods. The resulting CVS does not depend on process variables, overcoming a limitation present in ordinary models. In addition, the CVS was able to correctly classify the images from the test dataset uniformly, being able to extract the main colors present in the images from the dataset already in the first convolutional layer. The results showed a potential use of CNN in the food industry process control system involving color changes.

中文翻译：

---

短卷积神经网络应用于面包皮褐变阶段的识别

摘要 提出了一种基于卷积神经网络 (CNN) 的计算视觉系统 (CVS)，使用模块 Inception v3 和减少卷积层数 (Short-CNN) 进行操作，用于面包皮烘烤过程中的褐变程度分类。CNN 的训练 (70%)、验证 (15%) 和测试 (15%) 是使用由 374 个面包皮图像片段（600 × 600 像素）组成的数据集在七个烘烤周期内进行的。由此产生的 CVS 不依赖于过程变量，克服了普通模型中存在的限制。此外，CVS 能够从测试数据集中正确地对图像进行统一分类，能够从已经在第一个卷积层中的数据集中提取图像中存在的主要颜色。