A general purpose multi-fruit system for assessing the quality of fruits with the application of recurrent neural network

In the industry of agricultural farming, defected fruits are the major reason for financial calamities across the globe. It affects both the quality and competence of the fruits. Quality detection is a post-harvest process that requires highly skilled labor and time. Therefore, the automatic detection of quality of fruits is an important step in the harvesting process which helps to save the man power and time consumption. Different systems have been designed using image processing and learning techniques that detect the quality of fruits and classify them. To speed up the fruit sorting process, a system has been designed using machine learning and deep learning techniques. The proposed system works with total nine types of fruits: apple, banana, pear, guava, grape, mango, pomegranate, orange and tomato. Recurrent neural network is used as a proposed deep learning classifier which is trained using good and bad extracted features through implementation of principal component analysis. A simple contrast enhancement technique, preceded by grayscale conversion, has been used for balancing the unstable light in the input fruit image that can suppress the object definition. In the segmentation phase, canny edge detection is used for discovering the boundaries of the fruits. The comparative analysis with existing multi-fruit or single-fruit systems clearly shows that the proposed general purpose system overshadows them by achieving better accuracy (98.47%), precision (98.93%), recall (75.44%) and mean square error (1.53%) values.