Lawn Weeds Detection Methods Using Image Processing Techniques

概要

論文の詳細を見る
In this work, three methods of lawn weeds detection based on various image processing techniques, Bayesian classifier, morphology operators, and gray-scale uniformity analysis based methods, were evaluated and compared by using four different seasons image datasets. In the evaluations, two types of automatic weeding systems (i.e., chemical and non-chemical based) together with the detection methods were simulated and their performances were compared. From the results, for chemical approach, the Bayesian classifier based method could destroy 80.85%-96.30% of weeds, with more than 80% of accuracy for all datasets. For non-chemical approach, its accuracy was nearly 100% for all datasets. This shows its robustness against changing in season. The morphological operator based method was the best in weeds destruction for the non-chemical based system. However, its accuracy performance ranked as the last. For gray-scale uniformity analysis method, it missed detecting a lot of weeds for winter dataset, only 31.91%-36.17% of total weeds could be destroyed. Among three detection methods, the Bayesian classifier based method can be considered as the most appropriate method for both chemical and non-chemical weeding systems.
社団法人電子情報通信学会の論文
2006-10-13