Improved Faster RCNN for Ripeness and Size Estimation of Mangoes with multi-label output

Detecting fruit ripeness and size using deep learning is crucial for enhancing efficiency and accuracy in quality control within agriculture and food industries. Traditional manual methods are slow, labor-intensive, and prone to errors, while deep learning algorithms, like CNNs, automate the process, making it faster and more reliable. This paper dives into object detection methods in fruits, that is, detecting their size and ripeness. Various techniques in object detection were explored. Here, an object detection method using Faster-RCNN is proposed, where the results of 2 faster rcnn modules are combined, resulting in multiple outputs. A custom backbone is also proposed that enables to reduce the training time of the models. Through this, An accuracy of 75% and 63.34% on the individual ripeness and size is achieved and on combining, an accuracy of 58% is achieved. models respectively