[U-04] Geospatial Applications for Societal Benefits

[U04-04] Bagworm Detection in Oil Palm using Imaging Technology

*Siti Nurul Afiah Mohd Johari¹, Siti Khairunniza Bejo¹, Abdul Rashid Mohamed Shariff¹, Nur Azuan Husin¹, Mohamed Mazmira Mohd Masri², Noorhazwani Kamarudin² (1.Universiti Putra Malaysia, 2.Malaysian Palm Oil Board)

Keywords:Bagworm, Hyperspectral imaging, Machine learning, Deep learning, Unmanned aerial vehicle (UAV), Multispectral imaging

A serious outbreak of leaf-eating insects namely bagworm (Lepidoptera: Psychidae), especially Metisa plana species, may cause a 43% yield loss in oil palm production due to late proper control of bagworm populations. With a quick reproductive cycle and short lifespan, oil palms may eventually experience 10%–13% leaf defoliation. Identification and classification of bagworm instar stages are critical for determining the severity level of current outbreak before taking appropriate control measures in the infested area. A manual census was conducted to count the number of pests and determine the severity of infestation; however, when covering a large area, it typically takes more time and labor. Therefore, this study emphasizes the advancement of bagworm detection technologies, covering both identification of larval instar stage and evaluation their infestation severity levels. The capabilities of hyperspectral imaging combined with machine learning techniques were demonstrated to distinguish the bagworm larval instar stages, from second (S2) to fifth (S5) instar stages. Results have shown the dataset which used green bands at 506 nm and 538 nm with a weighted k-nearest neighbour classifier achieved the best value of accuracy (91% – 95%). The model also capable to detect the young larval instar stages (S2 - S3) where an active feeding activity takes place. These classifications were then upgraded to an automatic technique using transfer learning- based frameworks such as VGG16, ResNet50, ResNet152, DenseNet121, and DenseNet201. Among these models, DenseNet121 model, which used SGD with momentum (0.9) had the best classification which could identified all the instar stages from S2 to S5 with high value accuracy (94.52–97.57%). Meanwhile, machine learning technique was utilized to classify the severity levels of infestation in oil palm plantations based on the vegetation indices derived from the UAV imagery. Three types of combination among chosen vegetation indices were developed: NDVI-NDRE, NDVI-GNDVI, and NDRE-GNDVI. According to the results, the best combination in classifying healthy and low levels was found to be NDVI-GNDVI, with 100% F1 score. Weighted KNN become the best model that constantly gave the best performance in classifying all the infestation levels (F1 score > 99.70%) in all combination. These classifications method was then further extended by analyzing the impact of the dataset size on the machine learning performance. Resampling method of 3-interval undersampling was carried out and achieved 86.84% successful classification of 100% F1-score. Besides, fine KNN was constantly well performed in classifying all infestation levels in NDVI-NDRE combination across all datasets. The results clearly indicate that, even in cases where the data is unbalanced, there is a greater likelihood of success with the classification due to 66.67% reduction of the sample. The suggested technique is crucial for the early phase of severity-level detection and saves time on the preparation and operation of the control measure. These studies demonstrate that modern technology can precisely determine the bagworm's instar stage and severity level, which could ultimately be utilized to develop effective preventative measures to reduce and halt the bagworm's spread.