Recognition of Endemic Bird Species Using Deep Learning Models

Abstract

Numerous bird species have become extinct because of anthropogenic activities and climate change. The destruction of habitats at a rapid pace is a significant threat to biodiversity worldwide. Thus, monitoring the distribution of species and identifying the elements that make up the biodiversity of a region are essential for designing conservation stratagems. However, identifying bird species from images is a complicated and tedious task owing to interclass similarities and fine-grained features. To overcome this, in this study, we developed a transfer learning-based method using Inception-ResNet-v2 to detect and classify bird species endemic to Taiwan and to distinguish them from other object domains. Furthermore, to validate the reliability of the model, we adopted a technique that involves swapping misclassified data between training and validation datasets. The swapped data are retrained until the most suitable result is obtained. Additionally, fivefold cross-validation was performed to verify the predictive performance of the model. The proposed model was tested using 760 images of birds belonging to 29 species that are endemic to Taiwan; the images were captured from various environments with different perspectives and occlusions. Our model achieved an accuracy of 98.39% in the classification of the bird species and 100% in the detection of birds among different object categories. Moreover, the model achieved a precision, recall, and F1-score of 98.49%, 97.50%, and 97.90%, respectively, in classifying bird species endemic to Taiwan.

On the Feasibility of Privacy-Secured Facial Authentication for low-power IoT Devices - Quantifying the Effects of Head Pose Variation on End-to-End Neural Face Recognition

“…Deep Learning has been applied in soundscape ecology, zoology and ethology research projects were primarily interested in species identification [Selin et al, 2006, Chou et al, 2007, Sprengel et al, 2016, Lasseck, 2018a, Christin et al, 2018, Sankupellay and Konovalov, 2018, Lasseck, 2018b, Zhang et al, 2019, Koh et al, 2019, Ruff et al, 2020, LeBien et al, 2020, Ruff et al, 2021, Huang and Basanta, 2021, Campos Paula et al, 2022. Widely used algorithms in this context are Deep Neural Networks and Convolutional Neural Networks (CNNs) [Ruff et al, 2020, Christin et al, 2018, Zhang et al, 2019, Ruff et al, 2021, Hidayat et al, 2021, Kahl et al, 2021, Permana et al, 2021, Disabato et al, 2021.…”

Bird Species Identification Using Convolutional Neural Network

“…In this [8], a novel deep learning model was proposed to classify bird species along with another deep learning model using pre-trainedResNet architecture. The end-to-end deep network for fine-grained visual categorization [9] called Collaborative Convolutional Network (CoCoNet) was proposed and the implementation and performance of the model were based on the Indian bird's dataset. [10] A transfer learning-based method using InceptionResNet-v2 was developed to detect and classify bird species.…”

Refining manual annotation effort of acoustic data to estimate bird species richness and composition: The role of duration, intensity, and time

“…Limitations to automatic classification include generalizability to unmatched conditions, the availability of large previously annotated datasets, low accuracy, low robustness to noise such as wind and rain, the need for manual tuning of algorithm parameters, post‐processing of results, and sufficient expertise in machine learning (Stowell et al, 2019 ). However, increasingly sophisticated detection algorithms have demonstrated their ability to overcome many of these obstacles (Kahl et al, 2021 ; Stowell et al, 2019 ; Wood et al, 2021 ), and progress in this field is rapidly advancing (Denton et al, 2022 ; Huang & Basanta, 2021 ; Liu et al, 2022 ), particularly as more manually annotated files are added to existing datasets (Wood et al, 2022 ; Zhong et al, 2021 ) that researchers can use without building new algorithms. This method is particularly promising because once a classifier can produce robust results for a particular area, automated (and theoretically continuous) monitoring becomes possible, overcoming the current considerable limitation of time available for manual annotation alone to estimate which species occupy that area over time.…”