GRI Publications and Scholarship

Abstract

We present a dataset for the autonomous identification of invasive aquatic plant species using deep learning techniques. The dataset includes high-resolution images captured using a Canon EOS REBEL T3 DSLR camera for model training and low-resolution images captured with a Raspberry Pi camera for evaluating model performance under real-world conditions. Eight invasive aquatic plant species: Alternanthera philoxeroides, Cyperus blepharoleptos, Salvinia molesta, Ludwigia peploides, Panicum repens, Pontederia crassipes, Pistia stratiotes, and Nymphaea odorata were cultivated in mesocosms at the R.R. Foil Plant Research Center, Mississippi State University. The controlled environment of mesocosms serve as an intermediate between lab and field studies, providing insights into natural conditions before field testing. For training dataset, we captured 1,963 high-resolution images under natural lighting and various orientations, with 150–200 images per species. And to evaluate the models' robustness in real-world scenarios, a testing dataset of 128 images with lower resolution and color accuracy was created using a Raspberry Pi camera.

Publication Date

2025

Research Center

Geosystems Research Institute

Disciplines

Botany

Readme.pdf (2357 kB)

Included in

Botany Commons

Share

COinS