Evaluation of parameters in a neural network for detection of red ring pest in oil palm

Abstract

In Latin America, Ecuador, Colombia, and Honduras have the highest oil palm production, diseases that quickly spread in the plantations affect this production. That is why it is important for these countries to have control over the plagues. Using pesticides produces environmental pollution and respiratory problems for workers. Therefore, is necessary to apply precision farming and technologies that can prevent from having to use pesticides when the palm tree is already dead, to reduce the pollution and increase the quality of the palm oil. The aim of this project is to evaluate different neural networks for detection of diseases and plagues. This parameter starts with the type of device for data acquisition, type convolutional neural network, the weight trim rate, and false alarm rate. The design method was developed using incremental strategies. Finally, the drone Haar, network with a weight trim rate of 90% can detect the disease with a 98% of accuracy. In conclusion, the robot has a better accuracy when it uses an LBP training and when the positive and negative images are the same. However, its accuracy is low compared to the drone network using Haar training.