Deep neural network based fruit identification and grading system for precision agriculture

Precision agriculture is a systematic scientific way of farming with an aim to increase yield in terms of quality and quantity, ensure profitability, sustainability, and protection of the environment through optimum use of resources. Fruit quality is associated with several factors such as freshness level, vitamin content, nutritional content, etc. Grading of fruits is an important process as it ensures exact characteristics and helps in regulating price as per quality. With an increased demand for quality fruits, automated fruit grading systems are required to support the farmers and to maintain a sound pricing scheme in the market. This paper presents an automated fruit grading scheme based on deep neural networks. A convolutional neural network structure is proposed and compared with transfer learning-based models for grading the fruit images in the 'Kaggle' dataset. Remarkable results are achieved with the proposed as well as with the 4 pre-trained models, i.e., ResNet-50, VGG-16 & 19, and Inception-ResNet V2. Further the results are also compared with that obtained from the support vector machine classifier. For comparison of all these models, the input data is same i.e., intensities of the pixels in the input image. A user interface demonstration module is generated to demonstrate how it can be useful to the farmers for sorting their products.