UAV and a deep convolutional neural network for monitoring invasive alien plants in the wild

Abstract

Invasive alien plants (IAPs) are considered to be among the greatest global threats to biodiversity and ecosystems. Timely and effective monitoring is important for their prevention and control. However, monitoring remains mainly dependent on satellite remote sensing and manual inspection, which has a high cost and rather low accuracy and efficiency. We considered that this problem could be solved using unmanned aerial vehicle (UAV) intelligent monitoring. Accurate and rapid identification of IAPs in the wild is the core of intelligent monitoring. We intended to acquire colour images of the monitoring area in a field environment using an UAV and proposing a novel IAPsNet based on a deep convolutional neural network (CNN) to identify the IAPs appearing in the images. 6400 samples were one by one manually divided into seven IAP categories and one background category as training set. IAPsNet incorporated AlexNet local response normalization (LRN), GoogLeNet inception models, and continuous VGG convolution. Through training and testing, the IAPsNet performance for 893 testing samples was rather satisfactory, reaching an accuracy of 93.39% within a time of 1.8846 s and the average recall, average precision and average F1-score can reach 93.3%, 93.74% and 93.52% respectively. Moreover, in quantitative and qualitative comparative analysis, IAPsNet not only has high accuracy, high recall, high precision, high F1-score and efficiency but also has a high anti-interference capacity against blur, environment and multi-scales. Additionally, IAPsNet was applied to 4 different real wild conditions, proving that it is able to adapt to different scenes and simultaneously identify multiple species; it has potential to be used in the wild. High-quality distributional data of invasive plants are provided for subsequent ecological analysis. The data will help management authorities to implement the necessary steps in an identified area to develop a comprehensive strategy for IAP control.