The ASUCAL: Maturity analysis and monitoring system for sugarcane crops

• Main Authors: Camacho, Jonathan Clark S., Co, Anthea Marina A., Hao, Jan Percival S.J., Salazar, Ana Carmela P.
• Format: text
• Language: English
• Published: Animo Repository 2012
• Subjects: Sugarcane industry > Philippines; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/11542
• Institution: De La Salle University

The maturity period of sugarcane crops ranges from 8-to-12 months depending on the soil quality and amount of rainfall. Determination of crop maturity was initially based on a fixed number of months from first plantation and has become an unreliable method because of erratic number of months from first plantation and has become an unreliable method because of erratic weather conditions experienced in recent years. The crops must be harvested at the correct maturity level to achieve the needed consistency of sucrose content. To determine the correct maturity level, the crop is subjected to visual inspections on the stem node distances, and stalk and leaf color. Existing methods used in determining the maturity and projecting the sugar yield are through ocular inspection, refractometry and saccharometry. In this research, a new technique is introduced to determine crop maturity levels, using machine vision techniques. Regions of interest were identified in sugarcane images via thresholding in the Hue-Saturation-Value (HSV) color space, whereby 2-D HSV Histogram plots can be used as features in identifying a sugarcane crop's maturity. Two algorithms were compared to verify the best machine learning technique to be used in the system namely, Random Forest and Support Vector Machines (SVM). There are three data sets by which the best machine learning algorithm is verified: an unbalanced training set which is composed of 51 matured and 376 not matured sugarcanes, a balanced training set which is composed of 73 matured sugarcanes and 73 not matured sugarcanes, and monthly data sets of sugarcanes being monitored. Generally, SVM has a higher accuracy based on the model building and unbalanced training set. Random Forest is preferred over SVM algorithm for the system's machine learning algorithm because it exhibits a higher precision and recall, which is more reliable in classifying sugarcanes, matured or not matured in the monthly data sets. It also exhibits higher accuracy during Month 11 and Month 12 of the balanced data set. The developed system is 90.16% accurate. A focus group discussion was conducted to 10 sugarcane farming practitioners and agriculturists to benchmark the system it is revealed that the system is 29.66% more accurate and faster than the sugarcane experts.