Real-time oil palm fruit bunch ripeness grading system using image processing techniques

Fruits and other agriculture products are valued by their appearance, which is a major factor in the judgment of quality. The human eye, for example, has historically judged quality via appearances. External features and properties such as colour,texture, shape, and size are good indicators for parameters like ripeness and defects. Grading varies among graders and is often inconsistent. The adaptation of human eye to small changes in colour and the effect of the background on the perceived colour and intensity are the main sources of error. Hence, grading system technologies offer a solution to these problems. The grading systems in general utilized improved engineering designs with image processing techniques to ensure the quality of the product. In this research, a real time oil palm grading system was built and an image processing techniques algorithm was developed based on the external features of oil palm fresh fruit bunches (FFB) such as colour, texture, and thorns. The purpose of which was to investigate the relationship between the external features and ripeness of different oil palm FFB types as well as to test and validate the implementation of oil palm grading system methods and techniques. Special grading system with specific methods and techniques was built with fast, accurate, and objective ripeness classification to work with the parameters and properties of oil palm FFB, which is important for the farmers to have an objective classifier before selling their product as well as the oil palm companies to classify correctly the quality of oil palm fruit bunches due to the variations in different oil palm qualities. Image processing approaches, such as acquisition, pre-processing, segmentation,feature extraction, and classification as well as expert rule-based system, were developed to automate the ripeness grading for oil palm fruit bunches. Feature extraction for oil palm FFB colour, texture, and thorns was implemented by using statistical colour features, colour histogram, grey-level co-occurrence matrices (GLCM), basic grey level aura matrix technique (BGLAM), and Gabor wavelet techniques on the three different regions of interest (ROIs), namely, ROI1, ROI2, and ROI3. These ROIs were based on the training and the testing of the ANN, KNN, and SVM supervised machine-learning classifiers. Statistical measurements, such as the area under the receiver operating characteristic (ROC) curve (AUC), are used to evaluate classifier performance. The performance results showed that BGLAM, which was based on the ANN classifier and applied on the ROI3, was the optimal technique for grading oil palm FFB types with 93% performance accuracy and a 0.44 second processing speed. Furthermore, the grading system graded the oil palm FFB ripeness based on three different models. First, a significant 93% performance accuracy and a 1.6 second processing speed were achieved by combining the colour histogram and the ANN classifier applied on ROI3 based on the Nigrescens and Oleifera colour model. A 1.4 second processing time was achieved when the combination was applied on ROI2 for the Virescens colour model. Second, BGLAM and ANN applied on ROI3 achieved 92% accuracy and a 0.43 second processing time for the Nigrescens texture model. BGLAM and ANN achieved 93% accuracy applied on the ROI2 with a 0.40 second processing time for the Oleifera and Virescens texture models, which are the optimal results based on the texture model. Third, GLCM and ANN applied on the ROI1 achieved 87% accuracy and a 3.7 second processing time for the Nigrescens thorns model, whereas BGLAM applied on the ROI3 based on SVM achieved 91% accuracy and a 1.20 second processing time for the Oleifera thorns model as well as 88% accuracy and a 0.83 second processing time for the Virescens colour model. These results are optimal based on the thorns model. A new approach was developed using expert rules-based system. This system is based on three different ROIs that showed the best rule-based results, and were selected for further testing stages. For example, the rule-based ROIs for statistical color feature extraction with KNN classifier at 94% were chosen. The ROIs that indicated results higher than the rule-based outcome, such as the ROIs of statistical color feature extraction with ANN classifier at 94%, were used for further FFB ripeness testing. The results show that the texture models gives the best alogrithm result for oil palm FFB types and ripeness classification, where the BGLAM based on ANN with ROI3 gives a high accuracy 93% with shorter image processing time 0.44 (s) for FFB type recognition, whereas the alogrithm of BGLAM based on ANN and ROI3 with accuracy 92% and short processing time 0.43 (s) for Nigrescens, as well as the alogrithm of BGLAM based on ANN and ROI2 with accuracy 93% and short processing time 0.40 (s) for Oleifera and Virescens. The best rule-based and ROIs results were selected for further testing stages. This research has achieved its stated goal of developing a real time oil palm grading system for automated FFB types and ripeness classification. This system will be useful to the oil palm plantations in Malaysia and the rest of the oil-palm growing world. The results will benefit oil palm engineers, mills, managers, small holders, and enforcement agencies.

Preview PDF FK 2013 25R.pdf Download (1MB) | Preview

Actions (login required)

View Item