The rise of Deep Learning has been accompanied by a sharp increase in the use of Convolutional Neural Networks and the prevalence of teachable machines to do the work. Convolutional networks were inspired by biological processes CNNs use relatively little pre-processing compared to other image classification algorithm. The correlation between Convolutional Neural Network and Training models is significant in Image recognition. Food image recognition is a unique branch of image recognition that aims to predict the correct name of fruits with images as input. CNN's are a very effective class of neural networks, responsible for image classification and prediction. Teaching a machine to do something like Image recognition seemed very intriguing to us. Moreover, there are countless real-world applications of this concept. It is in light of these reasons that we decided to work on Image Classification and Prediction using different Convolutional Neural Network and testing their accuracy with real-time data on an application named Frooty based on Flutter UI development kit, which predicts the name of Fruits using images. we used a common dataset, Fruits-360 to train our models and the results showed variations in predicting the fruit images. Specifically, the number of epochs and the number of layers in models contributed towards their efficiency, and the analysis favoured the VGG16 model as it predicted the most test cases correctly. The results suggest that the accuracy of Image Recognition can be increased by merging layers from different CNN models and using a more varied and unbiased dataset.
Food image recognition is a unique branch of image recognition that aims to predict the correct name of fruits with images as input. CNN's are a very effective class of neural networks, responsible for image classification and prediction. The main objective is to determine how different models of CNN such as VGG16, ResNet50, InceptionV3, MobileNetV2 can be used to increase the efficiency of predictions in Food image recognition. The steps involved in our project are as follow,
We have considered a common dataset named Fruits-360 for our project which consists of 131 varieties of fruits and vegetables. The images for training and test cases are also included in the dataset.
With the use of keras imported with tensorflow in python code lab, we generated different model.tflite files by importing various different CNN’s that we are researching in this project and also by setting its training parameters such as the number of epochs, layers, steps, type of optimizer, training rate.
We designed a mobile application named Frooty with the help of a Flutter UI development kit using Dart Programming and provided a presentable layout for the application. The coding part is done using Android Studio and necessary files and images were included in the asset folder.
D. Merge models with Application and Predict test cases The different model.tflite files generated were merged one by one and provided with test cases to test the prediction accuracy of different models when collaborated with the flutter application. The graphs for accuracy and loss of data were plotted based on the analysis.
