CogKnife Project

COGKNIFE: FOOD RECOGNITION FROM THEIR CUTTING SOUNDS

Takamichi Kojima, Takashi Ijiri, Jeremy White, Hidetomo Kataoka, Akira Hirabayashi

Abstract

In this study, we present “CogKnife”, a knife device which can identify food. For this, a small microphone is attached to a knife, which records the cutting sound of food. We extract spectrograms from the cutting sounds and use them as feature vectors to train a classifier. This study used the k-Nearest Neighbor method (k-NN), the support vector machine (SVM) and the convolutional neural network (CNN) to verify differences of the classification methods. To evaluate the accuracy of our technique, we performed classification experiments with six kinds of foods (apples, bananas, cabbages, leeks and peppers) in a laboratory environment. From 20-fold cross validation, we confirmed high recognition accuracies, such as 83% with k-NN, 95% with SVM and 89% with CNN.

Data Set

all sound data (zip)

Publication

[author version (pdf)] [IEEE Xplore (not availabl yet)]


Takamichi Kojima, Takashi Ijiri, Hidetomo Kataoka, Jeremy White, Akira Hirabayashi. CogKnife: Food Recognition From Their Cutting Sounds, Proc. CEA2016: 8th Workshop on Multimediafor Cooking and Eating Activities, Seattle WA, 2016/7/15.




back to top