Identification of dorsal and vertical surface of rubber seeds using image processing approach

Natural rubber tree or known as Hevea brasiliensis is an important economic resource for the world and one of major plantation crops in Malaysia. To increase rubber production, planting method of rubber seeds must be improvised. Proper placement of seeds are important in order to increase the germination rate of rubber seeds. Rubber has two different surfaces which are dorsal and vertical. There are numerous darker mottles on the dorsal surface whereas only a straight valley at the centre of vertical surface. Vertical surface needs to be placed downward, attaching to the soil and dorsal surface needs to be placed on the top, facing to the sky. Current method of planting rubber seed is by growing the seedlings in a nursery. It needs many labors to plant the seeds one by one in a polybags. This caused high cost of production due to high labor intensity. To reduce the labor intensity and improving the production efficiency, it is necessary to use an automatic detection technology. This study was conducted to identify the dorsal and vertical surface of rubber seeds using image processing approach. There were 1600 images of dorsal and vertical surfaces at different positions acquired using SM-P605 of Samsung Galaxy Tab in RGB color format. Significant difference between dorsal and vertical surface can be seen clearly at the center of the seed. In this study, horizontal position of the rubber seed image was used as the reference. Therefore, after underwent all the image pre-processing steps, the orientation of the seed was identified. The seed was rotated into horizontal position based on the identified orientation. Then, canny edge detection was used to extract the important edge at the center of the seed in the horizontal based. From the center edge region, five features were extracted i.e. maximum length of x-axis, ratio of y-axis to x-axis, number of pixels inside edge region, maximum convolution and number of intersections. These features were used to develop a new prediction model using conditional statement method in identifying dorsal and vertical surface. Besides prediction model, support vector machine (SVM) and artificial neural network (ANN) were also used to classify dorsal and vertical surface. The result had shown that all the samples were successfully rotated into the horizontal positions with an average of error of 0.52%. The developed prediction model gave the most accurate result with 88.75% accuracy as compared to ANN (82.61%) and SVM (72.25%).

Preview Text FK 2017 119 - IR.pdf Download (1MB) | Preview

Actions (login required)

View Item