Development of an Auto Discrimination System of Mechanical Properties of Material and Qualities of Final Products for Porous Foods

本研究的目的是建立一种食品材料机械性能作为过程控制指标和最终产品质量作为多孔食品质量控制指标的自动判别系统。机械损耗切线和松弛时间是面包面团和蛋糕面糊工艺控制的适宜指标。在焙烤过程中，65 ~ 85℃的贮藏模量和损耗模量是较好的指标。“Good dough and batter”有很长的放松时间。对原始图像进行3*3矩阵的4邻域锐化预处理。新发展的数字图像处理算法；“多重图像二值化方法”可以从五个不同阈值的复合二值图像中获得每个空气单元的清晰图像。二维离散傅里叶变换（2D-DFT）方法可以通过细胞的频率和方向的功率谱来表征每幅面包屑图像。采用小波变换方法对锯齿形压缩力-变形曲线进行时频分析，可用于膨化玉米零食脆度的评价。应用于商品白锅面包的数字图像分析方法和商品膨化小吃的时频分析方法均适用于膨化食品的质量评价。利用从开顶面包的好、普通、坏三类面包屑颗粒图像中获得的二维dft功率谱的16个频率分量和16个方向分量进行图像识别。通过反向传播算法，利用三层神经网络学习实现面包屑颗粒的识别。

The aim of this study was to develop an auto discrimination system of mechanical properties of food material as indices of process control and qualities of final products as indices for quality control for porous foods.Mechanical loss tangent and relaxation time were suitable indices for bread doughs and cake batters in process control. In baking process, storage modulus and loss modulus at 65-85ﾟC were good indices. 'Good dough and batter' had a long relaxation time. Original images were pretreated with 4-neighbor sharping in 3*3 matrix. Newly developed digital image processing algorithms ; 'Multiple Image Binarization Method' made it possible to obtain a clear figure of each air cell from composite binary image with five different thresholds. Two dimensional discrete Fourier transform (2D-DFT) method made it possible to characterize each image of crumb by power spectrum of frequency and orientation for cells. Time-frequency analysis of jagged compressive force-deformation curves using the wavelet transformation method can be applicable to evaluation of crispness of puffed corn snacks.Digital image analysis methods developed for commercial white pan breads and time-frequency analysis method developed for commercial puffed snacks were applicable to quality evaluation of extruded foods. Sixteen frequency components andsixteen direetion components of 2D-DFT power spectra obtained from crumb grain images in 3 categories (i.e.good, ordinary, and bad) of open top type breads were used for image recognition. Crumb grains of the breads could be discriminated with three-layr neural network learning by back-propagation algorithm.