Transfer Learning Based Automated Crop Analysis

In our research paper, the technique of transfer learning has been applied to the classification of fruits, crops and vegetables. Every year the quality of large amount of crops and fruits deteriorates because of the lack of essential nutrients provided to them. In many areas, more than one crop are harvested in a single field which makes it very difficult for the farmer to provide proper nutrients to each and every type of crops as every crop requires specific kind of nutrients. To overcome this scenario, the transfer learning technique has been applied using pre-trained models for the segregation of crops based on the features like size, colour, quantity and so on. We have implemented two pre-trained deep learning models namely ResNet50 and MobileNet in order to differentiate between multiple types of crops, fruits and vegetables. To evaluate these models, we collected datasets for 15 different classes of fruits and crops that are referred as Multi CVF dataset, where CVF stands for crops, vegetables and fruits. The dataset is in two sets with automatically and manually collected samples from different sources. The dataset of each class consists of approximately 500 images. The experiment has also been extended for publically available Fruit 360 dataset. The result of the models shows that ResNet50 model achieved the higher test accuracy as compared to the MobileNet model.