Performance Evaluation and Application of Mask R-CNN with Common Objects in Context for Cacao Image Segmentation and Disease Classification

With the emergence of Artificial Neural Networks, the field of deep learning has taken a dramatic turn in recent years. In conventional machine learning tasks, these biologically inspired computational models exceed earlier types of artificial intelligence by a wide margin. The study aimed to create a Mask-RCNN model that can produce a performance metrics based on the data that has been gathered. On this research study, the researchers focused on the performance evaluation and application of Mask R-CNN with common objects in context for cacao image segmentation and disease classification. TensorFlow was used as a tool to help the AI model train based on the given data that has been gathered. The Mask-RCNN resulted with a training loss of 0.4057, a validation loss of 0.1159 an mAP of 0.712 (71.2%), an mAR of 0.829 (82%), and an f1-score of 0.766 (77%). On the other hand, the performance metrics of the model in terms of validation dataset resulted a recall of 0.504 (50%), an mAR of 0.738 (74%), and an f1-score of 0.599 (60%).