Integrated fuzzy and deep learning model for identification of coconut maturity without human intervention

Maturity of the coconut is judged manually based on the shape, color, weight, shaking sound, timeframe, etc. Currently, there is not much research being done to find solutions involving the latest techniques including fuzzy, machine learning and deep learning techniques for identifying maturity of the coconuts without human intervention. These techniques have significant challenges in detecting the multiple classes in coconut maturity due to difference in size and shape of the coconuts and the complex nature of their backgrounds as each coconut tree and each coconut bunch are unique. In this research work, we propose an integrated fuzzy and deep learning model (IFDM) to identify the maturity classes of coconut. Deep learning-based Mask R-CNN is used to classify and segment the coconut images. The combined fuzzy system (CFS) extracts the color, shape and scratch features with a fuzzy extraction technique for each feature. A probability-based fuzzy integration (PFI) method integrates the output from each fuzzy extraction to classify the coconuts. Decision-making probability (DMP) model combines the class probability of the probability-based fuzzy integration method and Mask R-CNN model to obtain the final coconut maturity class. The dataset with 10,000 images is used for training. Two test datasets, the first one containing 1754 real-time coconut images and the second containing 1000 cropped coconut images from the background, are used for testing. Using the proposed learning model in real time, an accuracy of 86.3% is obtained in classifying the maturity classes of coconuts. While integrating the models like Yolo v3, Yolo v5, Faster R-CNN, random forest and KNN with combined fuzzy system, Mask R-CNN performed better in terms of accuracy.