The project’s goal is to identify a tree as infected with the “Red Palm Weevil“, based on recordings of the trunk.
Red Palm Weevil is a beetle that nests in a tree (for several cycles) and feeds of the inside of the trunk until the tree dies and falls. The idea behind this project is to use recordings of the palm trunk (taken by the farmer) to detect infestations of tree as early as possible in the process. The assumption is that the sounds of the feeding can be discerned. Such deduction will enable the farmer to handle the palm or prevent further infestations.
In this project we used recordings made by Prof. Amotz Hetzroni and Daniel Katz - of infected and healthy trees. We tested different methods to create a solution and eventually based the system on the article “Detection of the red palm weevil rhynchophorus ferrugineus using its bioacoustics features”.by alid Barakat Hussein, Mohamed Ahmed Hussein & Thomas Becker.
Each full record is divided into samples of variable sizes and overlaps. For each sample, time and frequency features are extracted. This process gives a vector of samples with features, classified into 2 classes - infected palm / healthy palm. Based on this vector, we generated a training group and test group - these groups were used to train and create different classifiers (KNN, decision trees, SVM). After examining the classifiers, we chose the best classifier.
The Boosted Trees classifier with 30 Trees with depth of 6 and built the system according to it: for new recording – divide to samples -> extraction features -> decision making - whether the tree is infected or not.
the system identifies the recordings in 95% success rate with random trainings.
Daniel Katz, R&D Eden Farm, Emek HaMayanot