Fuzzy classification of pre-harvest tomatoes for ripeness estimation – An approach based on automatic rule learning using decision tree

• Red-green color difference has the good classification capability than that from individual components R, G, B or red-green ratio in RGB color space.
• Adaptive segmentation technique segments the unripe tomato images from the background with a good effect.
• Decision trees for automatic rule learning repudiate the need of a human expert.
• Automatic fuzzy partitioning of the feature space to over linguistic terms.
• High true positive rate and lower false positive rate for the proposed fuzzy rule base classification. High classification accuracy achieved by the proposed system over learning algorithms.

Tomato (Solanum lycopersicum) ripeness estimation is an important process that affects its quality evaluation and marketing. However, the slow speed, subjectivity, time consumption associated with manual assessment has been forcing the agriculture industry to apply automation through robots. The vision system of harvesting robot is responsible for two-tasks. The first task is the recognition of object (tomato) and second is the classification of recognized objects (tomatoes). In this paper, Fuzzy Rule-Based Classification approach (FRBCS) has been proposed to estimate the ripeness of tomatoes based on color. The two color depictions: red-green color difference and red-green color ratio are derived from extracted RGB color information. These are then compared as a criterion for classification. Fuzzy partitioning of the feature space into linguistic variables is done by means of a learning algorithm. A rule set is automatically generated from the derived feature set using Decision Trees. Mamdani fuzzy inference system is adopted for building the fuzzy rule based classification system that classifies the tomatoes into six maturity stages. Dataset used for experiments has been created using the real images that were collected from a farm. 70% of the total images were used for training and 30% images of the total were used for testing the dataset respectively. Training dataset is divided into six classes representing the six different stages of tomato ripeness. Experimental results showed the system achieved the ripeness classification accuracy of 94.29% using proposed FRBCS.

Figure optionsDownload as PowerPoint slide