糖熏鸡腿颜色快速精准识别的多层卷积神经网络模型研究

Vision research on CNN model for quick and accurate identification of sugar-smoked chicken thighs color

为快速精准识别糖熏鸡腿在熏制过程中产生的所有颜色,基于机器视觉技术,构建Xception-CNN模型用于熏鸡腿颜色的识别,同时应用Res Net-50、Inception和传统卷积神经网络(convolutional neural networks,CNN)等3种模型对比分析Xception-CNN模型对熏鸡腿颜色的识别效果。采集并经过图像预处理后,共得到不同颜色的熏鸡腿图像4 352张,作为4种模型的实验样本,随机选取其中的3 482张作为训练组,剩下的870张作为测试组。结果表明,4种模型的平均识别准确率分别为92%(Xception-CNN)、91%(Res Net-50)、89%(Inception)、87%(传统CNN);测试时间分别为1. 36 s(Xception-CNN)、0. 81 s(Res Net-50)、0. 98 s(Inception)、2. 48 s(传统CNN)。Xception-CNN模型对糖熏鸡腿图像的颜色识别准确率最高,达到92%,测试时间略高于Res Net-50模型和Inception模型,但低于传统CNN模型,仅需1. 36 s即可完成识别,此模型可以实现糖熏鸡腿颜色的快速精准识别,为糖熏工艺参数精准调控、保障产品颜色标准化等提供可靠依据。

To explore the machine vision model that can quickly and accurately identify all the colors of smoked chicken legs during the smoking process,this article builds an Xception-CNN model for the recognition of smoked chicken leg colors based on machine vision technology,while compares and analyzes the Xception-CNN model’s recognition effect on smoked chicken leg color by applying ResNet-50,Inception and traditional CNN and other three models.A total of 4352 smoked chicken leg images of different colors were collected and image pre-processed as experimental samples of four models,of which 3482 were randomly selected as the training group,and the remaining 870 as the test group.The average recognition accuracy of the four models is Xception-CNN-92%,ResNet-50-91%,Inception-89%,traditional CNN-87%,and test time is Xception-CNN-1.36 s,ResNet-50-0.81 s,Inception-0.98 s,traditional CNN-2.48 s.Therefore,the Xception-CNN model has the highest color recognition accuracy for sugar-smoked chicken leg images,reaching 92%.The test time is slightly higher than the ResNet-50 model and the Inception model,but lower than the traditional CNN model.It only takes 1.36 s to complete the recognition.The fast and accurate identification of the color of smoked chicken legs can provide a reliable basis for the precise control of the parameters of the sugar smoked process and the standardization of product colors.