【作者】 任珂；

【导师】 屠康；

【作者基本信息】 南京农业大学 ， 食品科学， 2006， 硕士

【摘要】 本文以青花菜为试材，建立青花菜采后颜色变化动力学模型，提出新的黄化分级标准，利用计算机视觉技术提取青花菜的颜色及形状特征参数，用统计模式与神经网络对特征参数进行分级，同时采用半导体激光技术控制青花菜采后病虫害(果蝇)。1．通过建立青花菜采后颜色变化动力学模型，确定了与人工分级指标相关联的3个颜色特征参数(b、TCD、H°)，并给出了其在各级别中的阈值；动力学模型显示，b和TCD符合Arrhenius一级动力学反应模型，而α和H°合二项式模型；同时根据青花菜实际商品性状的变化，提出了以黄化面积比为标准的4等级新分级标准。2．利用计算机视觉图像分析技术对320幅青花菜图像进行特征值提取及分析，通过背景纯化、颜色分割、灰度转换等图像分析手段从图像中提取出5个颜色及形状特征值(b、TCD、H°、黄化面积比、圆形度)；采用统计模式(多元线性回归)对5个特征值进行建模，并预测分级，预测精度仅达56.9％，效果不太理想；通过VC++6.0的程序设计建立青花菜特征值的自动提取及分级系统。3．通过MATLAB7.0建立5种神经网络(3层前向传递BP神经网络、概率神经网络PNN、自组织竞争神经网络SOC、学习矢量量化神经网络LVQ、自组织特征映射神经网络SOM)，采用320幅图像特征值训练网络并进行自回归验证，100幅图像特征值进行预测验证，结果表明，4种分类神经网络(PNN、SOC、LVQ、SOM)与BP神经网络均可进行青花菜分级处理(预测精度在68.2～93.4％之间)，但比较发现，BP神经网络最佳，预测精度高达93.4％；PNN与BP的预测精度及验证精度均相差不大，且运算时间只需BP的1／5，因此也有一定实用价值。4．利用半导体激光技术对青花菜采后病虫害(果蝇)进行控制研究，采用响应曲面法设计，研究激光功率和照射时间对果蝇的致死等生物学效应。结果表明，用波长范围650nm的半导体激光照射果蝇幼虫，在功率60mW、照射1282s条件下，果蝇幼虫的致死率可达到99％以上，且体重降低、羽化延续时间缩短，激光对果蝇幼虫具有较强生物学致死效应。但当功率低于40mW时，激光对果蝇产生激励作用，促使果蝇的生长发育。继续用原剂量照射第三代果蝇的幼虫，对比发现在连续多代照射后，果蝇后代并没有产生对激光的抗逆性。更多还原

【Abstract】 The kinetic models of post-harvest broccoli color changes were developed in this paper. And the new grading standard based on yellowness was established. The feature parameters of broccoli color and shape were extracted by computer vision and were graded by the statistical mode and artificial neural network. The insect pests (drosophila melanogaster) of post-harvest broccoli were controlled by the semi-conductor laser technique at the same time.1. According to the developed kinetic models of post-harvest broccoli color changes, the three color feature parameters (b~*, TCD, H~°) associated with artificial grading index were confirmed and their threshold in each class were given out respectively. The kinetic models showed that the rate constants of color parameters b~* and TCD were following a first-order Arrhenius-type reaction, and the polynomial model was suitable for the changes of a~* and H~°value. Meanwhile, on the basis of actual merchandise value changes of broccoli, the new grading standard was established with 4 classes based on the proportion of yellowness area.2. The feature parameters of 320 broccoli images were extracted and analyzed with computer vision analysis technique. The five feature parameters of color and shape (b~*, TCD, H~°, yellowness area proportion, roundness) were extracted from those images by the image analysis method, such as background purification, color segmentation, gray transform etc. The statistical mode (Multiple Linear Regression) was used to develop the model of grading based on those feature parameters, with undesired result at a accuracy of 56.9%. The automatic extraction and grading system based on the feature parameters of broccoli was developed by the program design in Visual C++ language.3. Five artificial neural networks (3-Layer Back Propagation Neural Network, Probabilistic Neural Network, Self-Organizing Competition Neural Network, Learning Vector Quantization Neural Network, Self-Organizing Feature Map Neural Network) were used as classifier in MATLAB7.0. The feature parameters of 320 images were used to train these networks and self-regression validity test, while that of 100 images were used for forecasting Validity test. The results showed that 4 kinds of classify neural networks (PNN, SOC, LVQ, SOM) and BP neural network were suitable for broccoli grading with the forecasting accuracy at the range of 68.2-93.4%. By contrast of these networks, BP neural network was the best network with the forecasting accuracy at 93.4%. PNN has certain application value since the difference of forecasting accuracy and self-regression validity between PNN and BP network were not great, and provided with the only 1/5 of running required time of BP.4. In order to study the control effects of the main insect pests (drosophila melanogaster) on post-harvest broccoli with the semi-conductor laser technique, the experiment designed with response surface investigated the biological effects of drosophila melanogaster for different laser power and irradiation time, such as dead. The results showed that the death rate was above 99% with the condition of laser power of 60 mW and irradiation time of 1282s in the wave-length of 650nm when the larva of drosophila melanogaster had been dealt with semi-conductor laser. While the weight of drosophila melanogaster was reduced, the required eclosion time was decreased. Therefore, the semi-conductor laser has strong biological effects to larva of drosophila melanogaster. However, when the power was below 40 mW, the laser light has the effect of promoting the growth of drosophila melanogaster. The experiment disclosed that drosophila melanogaster did not cause anti-laser effects on the third generation under the same experimental conditions in contrast to comparative groups.更多还原