Automated Mango Fruit Recognition by Multi-Task Convolutional Neural Networks for Harvest Robot

Efficient and effective mango fruit recognition is viewed as significant for development of a smart agriculture robot (ARo) for yield prediction, pest control, sorting and fruit detection. Several fruit recognition techniques for structuring ARo have been employed during the most recent decades. Recently, ordinary natural fruit identification techniques are lacking progressive response, exactness and extensibility. In this paper, we proposed an improved algorithm of MTCNN (Multi-Task Cascaded Convolutional Network) based on IFD (Intelligence Fruit Detection) technique. This technique has the ability to make the ARo work progressively with high precision. Additionally, in view of the connection between the number of tests on dataset and the boundaries of Neural Networks advancement, this work presents an improved strategy. A method that depends on image clustering is used to improve the identification in this project. The experimental results exhibited that the proposed identification performed significantly, both as far as exactness and time-cost. Besides, the broad trials exhibited that the proposed strategy has the limit and a decent compactness to work with other associated systems.