Precision monitoring of leaf-cutting ant nests in sub-orbital RGB images using deep learning
Leaf-cutting ants are the main pests of plantation forests in South America, causing severe defoliation, leading to production losses, plant mortality and increased susceptibility to other insects. Chemical control with sulfluramid active ingredient is the most used method. This proposal aimed to develop an innovative method for leaf-cutting ants nest detection in sub-orbital RGB images using deep learning techniques. The study was carried out in a 6-month-old eucalyptus stand with 91.3 hectares in the municipality of Três Lagoas, in the state of Mato Grosso do Sul, Brazil. The images of the stand were collected by a DJI Phantom 4 Advanced aircraft with an RGB camera and processed to produce an orthomosaic with a ground-level resolution of 5.2 cm/pix. The final orthomosaic was cropped in sub-images of 98 x 81 pixels. Sub-images that contained ant nests were labelled using bounding boxes. The database used in experiments consists of 2465 images containing leaf-cutting ant nests and 2465 images of ants' nest absence (background). The detection algorithm used as the deep learning framework based on the YOLO convolutional neural network architecture. The quality of its predictions was evaluated by accuracy, Kappa, sensitivity, specificity, and absolute mean error (MAE) metrics between training and validation samples. YOLO achieved 98.45% accuracy and YOLO 0.49% MAE as the best performances in nests measuring task, demonstrating the high complexity of detecting this target type. Obtained results show that YOLO is a promising approach for precision monitoring of leaf-cutting ant nests in sub-orbital RGB images and can contribute to reduce and optimize insecticide use in plantation forests, which is aligned with the UN Sustainable Development Goals (SDGs), consisting of responsible consumption and production (Goal 12), and terrestrial life (Goal 15). Keywords: Monitoring and data collection, Innovation, Knowledge management ID: 3622337
| Main Author: | |
|---|---|
| Format: | Document biblioteca |
| Language: | English |
| Published: |
FAO ;
2022
|
| Online Access: | https://openknowledge.fao.org/handle/20.500.14283/CC2646EN http://www.fao.org/3/cc2646en/cc2646en.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|