A UAV and Deep Transfer Learning Based Environmental Monitoring: Application to Native and Invasive Species classification in Southern regions of the USA

Unmanned aerial vehicle (UAV) applications can be powerful tools in horticultural research. However, they have not yet been widely explored in the literature. One significant area of interest for the horticultural community is identifying invasive plant species that, if left unchecked, can hinder the growth and health of native plant species. To address this issue, we assembled a novel data set of invasive and native plant species for seven southern states in the United States and developed a plant classification technique using pre-trained convolution neural networks and transfer learning. We explored extracting features from our data set using several state-of-the-art deep convolution neural network models, including InceptionV 3, MobileNetV 2, ResNetV 2, VGG16, and Xception. We then used the extracted features to classify the plant species using a convolutional neural network with cross-validation. Our experiments demonstrated the potential of our proposed method for achieving performance with a 94% accuracy using the MobileNetV2-DCNN model with data augmentation, hyper-parameter optimization, and the softmax classification technique. The advantage of our approach is the learning process is automated and highly accurate. The data set to train the plant species classifier will be available on request.