Identifying Epiphytes in Drones Photos With a Conditional Generative Adversarial Network (C-Gan)

Abstract

Abstract. Unmanned Aerial Vehicle (UAV) missions often collect large volumes of imagery data. However, not all images will have useful information, or be of sufficient quality. Manually sorting these images and selecting useful data are both time consuming and prone to interpreter bias. Deep neural network algorithms are capable of processing large image datasets and can be trained to identify specific targets. Generative Adversarial Networks (GANs) consist of two competing networks, Generator and Discriminator that can analyze, capture, and copy the variations within a given dataset. In this study, we selected a variant of GAN called Conditional-GAN that incorporates an additional label parameter, for identifying epiphytes in photos acquired by a UAV in forests within Costa Rica. We trained the network with 70%, 80%, and 90% of 119 photos containing the target epiphyte, Werauhia kupperiana (Bromeliaceae) and validated the algorithm’s performance using a validation data that were not used for training. The accuracy of the output was measured using structural similarity index measure (SSIM) index and histogram correlation (HC) coefficient. Results obtained in this study indicated that the output images generated by C-GAN were similar (average SSIM = 0.89–0.91 and average HC 0.97–0.99) to the analyst annotated images. However, C-GAN had difficulty to identify when the target plant was away from the camera, was not well lit, or covered by other plants. Results obtained in this study demonstrate the potential of C-GAN to reduce the time spent by botanists to identity epiphytes in images acquired by UAVs.

The Effect of Contrast Enhancement on Epiphyte Segmentation Using Generative Network

“…Conditional GAN is widely used for many image generation applications. Shashank et al, (2020) used CGAN for identifying epiphytes from drone photos where the CGAN will generate the output label for input image. Jiao et al, (2019) evaluated the potential of the CGAN for plant leaf recognition.…”

“…This study addressed the image data generation case studies in synthesizing photos from label maps, reconstructing objects from edge maps, and colorizing images. The potential of such an image-to-image translation model for epiphyte segmentation with less training samples are studied by our research team (Shashank et al, 2020. The proposed study focused on the impact of quality of training data in terms of image contrast while training a CGAN network is addressed.…”

“…The conditional generative adversarial network architecture implemented in earlier study (Shashank et al, 2020) is used to train the new models with images enhanced by four different methods and a model without any enhancement. The input to the architecture is a pair of images with input image and corresponding annotated image.…”

“…Both the network have the conditional parameter as the real input image depicted as 'y' in Figure 3 and the corresponding label. The training and details of the CGAN network used for segmentation is given in (Shashank et al, 2020).…”

Learning and Adaptation From Minimum Samples With Heterogeneous Quality: An Investigation of Image Segmentation Networks on Natural Dataset

“…The input images hold a dimension of 512 × 512 × 3. The current study concentrates on a single epiphyte species named Warahuai Kupperina captured from the Costa Rica reserve forest [4], [5]. The quality factors of the epiphyte images, like lighting conditions, the distance at which the drone captures the images, and the occupancy of the target and background, vary throughout the image samples.…”