An Analysis of the Rice Leaves Color Intensity with Support Vector Regression to Reduce Production Costs

Abstract

Abstract

Farmers extensively plant the rice hybrids which are not sensitive to the light by selecting to fertilize the Nitrogen
fertilizer in the paddy fields to increase the productivity. Nevertheless, the Nitrogen fertilizer (Urea) is easily lost in the flooding
conditions. Thus, the using of the Nitrogen fertilizer in the paddy fields where are in the flooding condition for the whole year has
the efficiency worth with its amount. Fertilizing the Nitrogen fertilizer directly with the rice’s demand was recommended to farmers by applying the colorimetric plate of rice leaves as a tool for determining the fertilizer requirements of the paddy fields. The problem was the farmers need to choose the sample rice leaves at least 10 leaves throughout the paddy fields comparing with the Leaf Color Chart (LCC) that could measure only one leaf at a time. Moreover, the color of the rice leaves might change owing to the reflection of sunlight to the eyesight the observers. Furthermore, if the leave color obtained was ambiguous between two
color bands it would be difficult to decide the right crop nutrient requirements. As a consequence, an image processing-based technique was developed to solve these problems. Two colorimetric plates of rice leaves which had four highlight colors were scanned as image files. After that, the rice leaves were extracted from the image by segmentation using Fuzzy C-mean. The average RGB colors of the extracted rice leaves were then used as features for the SVR which determines the amount of Nitrogen fertilizer added in the paddy fields. From the experimental result, the average absolute error and mean absolute percentage were used as a criterion of the accuracy of the system. It also found that the method of Support Vector Regression yielded the sum of average absolute error and the percentage for the lowest average absolute error in the blind data compared with reading the rice color leaves from the LCC of the experts.