Automated crocodile detection using deep learning and synthetic data

This is the authors' abstract. We don't add key points for this paper.

Image object detectors are a powerful tool that allows researchers and farmers to automatically detect and analyse animals in aerial images using unmanned aerial vehicles. This technology reduces the cost and time required to do animal surveys. However, one major challenge of object detectors is the large amount of data required to train the model. This can especially become problematic if the target animal is scarce and there is no available data to train on. In this paper, the generation and use of synthetic data for training a “you only look once” (YOLO) object detector to identify crocodiles in aerial images were explored. Additional methods for improving the performance of the detector were also presented. The results demonstrated that the accuracy of nearly all the models was improved when the backbone layers were frozen. Further, fine-tuning on a small number of real-world images after initially training on a synthetic dataset, and using mixed datasets of synthetic and real-world images, further improves the model accuracy when compared to a model trained on only that small number of real-world images. With a reduced real-world dataset of 20% to 1% of the full dataset, an increase of accuracy is observed of 3.1% to 45.38% for fine-tuning, and 2.9% to 47.12% for mixed datasets. These findings highlight the potential for the use of synthetic data to enable effective wildlife monitoring in data-limited environments, supporting more efficient and scalable conservation efforts. • Study addresses limited real-world data for wildlife object detection models. • Synthetic data is used to improve aerial crocodile detection using YOLO. • Training on only synthetic data achieved 72.4% of full-data model accuracy. • Fine-tuning with 1% of real data achieved 82% of full-data model accuracy. • Mixing synthetic data with 1% of real data achieved 83% of full-data model accuracy.