摘要
【目的】增强荔枝表皮缺陷提取效果,满足其品质检测分级准确性要求。【方法】采用Tensorflow框架构建基于AlexNet的全卷积神经网络AlexNet-FCN,以ReLU为激活函数,Max-pooling为下采样方法,Softmax回归分类器的损失函数作为优化目标,建立荔枝表皮缺陷提取的全卷积神经网络模型,并用批量随机梯度下降法对模型进行优化。【结果】模型收敛后在验证集上裂果交并比(IoUd)为0.83,褐变交并比(IoUb)为0.60,褐变与裂果的总体交并比(IoUa)为0.68;与利用线性SVM、朴素贝叶斯分类器缺陷提取效果相比,该模型的特征提取能力显著提高。【结论】全卷积神经网络在水果表面缺陷提取中具有良好的应用前景。
【Objective】To enhance the effects of litchi fruit pericarp defect extraction and satisfy the accuracy requirements of quality detection and classification.【Method】A fully convolutional neural network was built up based on AlexNet(AlexNet-FCN) using Tensorflow framework, with ReLU as the activation function, Maxpooling as the down-sampling method and loss function of Softmax regression classifier as the optimization target. Mini-batch stochastic gradient descent(Mini-batch SGD) was used to optimize the model.【Result】When the model was converged, the intersection-over-union of dehiscent area(IoUd) of litchi fruit cracking was 0.83 for the validation set, the intersection-over-union of brown area(IoUb) was 0.60, and the intersection-over-union of both dehiscent and brown area(IoUa) was 0.68. Compared with linear-support vector machine(SVM) and Na?ve Bayes classifier, AlexNet-FCN had a stronger defect extraction ability.【Conclusion】Fully convolutional networks(FCN) have a good prospect for application of fruit pericarp defect extraction.
引文
[1]杨松夏,陆华忠,吕恩利,等.球形水果机械式大小分级技术研究现状[J].农机化研究,2012,34(8):235-239.
[2]赵茂程,侯文军.我国基于机器视觉的水果自动分级技术及研究进展[J].包装与食品机械,2007(5):5-8.
[3]饶秀勤.基于机器视觉的水果品质实时检测与分级生产线的关键技术研究[D].杭州:浙江大学,2007.
[4]CEN H Y,LU R F,ZHU Q B,et al.Nondestructive detection of chilling injury in cucumber fruit using hyperspectral imaging with feature selection and supervised classification[J].Postharvest Biol Tec,2016,111:352-361.
[5]熊俊涛,邹湘军,刘念,等.基于机器视觉的荔枝果实采摘时品质检测技术[J].农业机械学报,2014,45(7):54-60.
[6]杜永忠,平雪良,何佳唯.圣女果表面缺陷检测与分级系统研究[J].农业机械学报,2013,44(S1):194-199.
[7]赵娟,彭彦昆,SAGAR D,等.基于机器视觉的苹果外观缺陷在线检测[J].农业机械学报,2013,44(S1):260-263.
[8]李江波,饶秀勤,应义斌.基于照度-反射模型的脐橙表面缺陷检测[J].农业工程学报,2011,27(7):338-342.
[9]MILLER B.Peach defect detection with machinevision[J].Trans Asae,1991,34(6):2588-2597.
[10]田杰,韩冬,胡秋霞,等.基于PCA和高斯混合模型的小麦病害彩色图像分割[J].农业机械学报,2014,45(7):267-271.
[11]LEEMANS V,MAGEIN H,DESTAIN M F.Defect segmentation on‘Jonagold’apples using colour vision and a Bayesian classification method[J].Comput Electron Agr,1999,23(1):43-53.
[12]MIZUSHIMA A,LU R.An image segmentation method for apple sorting and grading using support vector machine and Otsu’s method[J].Comput Electron Agr,2013,94:29-37.
[13]YANG W,JIN L,TAO D,et al.Dropsample:A new training method to enhance deep convolutional neural networks for large-scale unconstrained handwritten Chinese character recognition[J].Pattern Recogn,2016,58(4):190-203.
[14]ALBU R D.Human face recognition using convolutional neural networks[J].JEEE,2009(2):110-113.
[15]RAMAIAH N P,IJJINA E P,MOHAN C K.Illumination invariant face recognition using convolutional neural networks[C]//IEEE International Conference on Signal Processing,US:IEEE Xplore,2015:1-4.
[16]SINGH R,OM H.Newborn face recognition using deep convolutional neural network[J].Multimed Tools Appl,2017,76(18):1-11.
[17]高震宇,王安,刘勇,等.基于卷积神经网络的鲜茶叶智能分选系统研究[J].农业机械学报,2017,48(7):53-58.
[18]马晓丹,关海鸥,祁广云,等.基于改进级联神经网络的大豆叶部病害诊断模型[J].农业机械学报,2017,48(1):163-168.
[19]谭文学,赵春江,吴华瑞,等.基于弹性动量深度学习神经网络的果体病理图像识别[J].农业机械学报,2015,46(1):20-25.
[20]LONG J,SHELHAMER E,DARRELL T.Fully convolutional networks for semantic segmentation[J].TPAMI,2014,39(4):640-651.
[21]KRIZHEVSKY A,SUTSKEVER I,HINTON G E.ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 15th International Conference on Neural Information Processing Systems.Cambridge:Mit Press,2012:1097-1105.
[22]陈石榕.国际鲜荔枝标准[J].标准化信息,1997(4):11-12.
[23]周先治,陈阳,唐建阳,等.荔枝褐变损失估计研究[J].热带作物学报,2012,33(8):1403-1408.
[24]ZEILER M D,KRISHNAN D,TAYLOR G W,et al.Deconvolutional networks[C]//Computer Vision and Pattern Recognition.US:IEEE Computer Science,2010:2528-2535.