ABSTRACT
This chapter analyzes the feasibility of using transfer learning to identify the changing process of fresh banana and establishes the relationship between fruit quality and time. It utilizes a pre-trained GoogLeNet model to identify the process of banana change. Transfer the last layer to the new classification task by replacing them with a layer of full connection, a softmax layer, and a classification output layer. Transfer learning is a new machine learning method which uses existing knowledge to solve different but related problems. It relaxes the two basic assumptions in traditional machine learning and aims to transfer the existing knowledge to solve the learning problem with only a small number of labeled sample data or even none in the target field. The GoogLeNet pretrains the ImageNet dataset sufficiently. Although the neural network before training cannot recognize the banana, it provides good initial values for the banana recognition network.
