Grading Of Pork Freshness Fusing Image And Olfactory Features

Color features and texture features,as important features of images,are an important aspect of image classification and recognition,and have been widely used in practical production and real life.Olfactory feature is an important feature to commonly distinguish different objects and has high reliability.Pork freshness is an important index of pork quality,and its freshness detection and grading is of great practical value to ensure the quality and safety of pork and safeguard the vital interests of consumers.Therefore,it is of great practical significance to apply image features and olfactory features to pork freshness grading.This paper firstly constructs the image acquisition system,which consists of digital camera,light box and upper computer.The Nikon D7000 digital camera take images of the pork samples,which placed in the light box.The gathered images are transmited to the PC through USB2.0 port.Based on Qt,the pork freshness grading software system is developed.A method of pork freshness grading fusing image and olfactory features is proposed.In the aspect of image processing,to obtain a target image which can be used in the color feature extraction,we processed the ROI of color feature to gray image,denoising image,segmentation image,morphology image,mask image in order.To obtain the target image which can be used for texture feature extraction,we processed the ROI of texture feature to gray image and denoising image.In the aspects of features extraction and features dimension reduction,color features extracted 22 dimensions of feature parameters in total,including the average grayscale,the R,G,B three components of RGB color model,the R,G,B three parameters of rg chroma space,H,S,V three components of HSV color model,L,a,B three-component of Lab color model,the first,second and third order color moments.The texture feature parameters were extracted eight dimensions,including the four most important feature vectors,which are energy,entropy,moment of inertia,correlation's mean and standard deviation in the Gray-level co-occurrence matrix.The olfactory features were extracted of 10 dimensions.Respectively,we used principal component analysis to reduce the dimension of color features,texture features,color and texture fusion features,olfactory features,image and olfatory fusion features.The color features were reduced from 22 dimensions to 7 dimensions,texture features from 8 dimensions to 4 dimensions,color and texture fusion features from 30 dimensions to 10 dimensions,olfactory features from 10 dimensions to 2 dimensions,image and olfatory fusion features from 40 dimensions to 10 dimensions.For the grading model construction,the 165 sets of data collected from 165 original images were randomly selected of 90 sets as the training set and 75 sets as the test set.In turn,by using the extreme learning machine model,random forest model,LVQ neural network model,support vector machine model,and 7 dimensions color features,4 dimensions texture features,10 dimensions color and texture fusion features,2 dimensions olfactory features,10 dimensions image and olfatory fusion features,whose dimensions were reduced,achieve freshness grading respectively.The research results show that the classification accuracy of the image and olfatory fusion feature is higher than the color feature,the texture feature,the color and texture fusion feature and olfactory feature in the four classification models.Whether the color feature,texture feature,color and texture fusion feature,olfactory feature or image and olfatory fusion feature,the support vector machine model is the best choice.