All Malayemys species are native freshwater turtles that are endemic to mainland Southeast Asia. Based on morphological and molecular characteristics, including microsatellite DNA and mitochondrial DNA, it was revealed that there are three species of snail-eating turtle, (1) Malayan snail-eating turtle M. macrocephala distributed in northern, central, and southern Thailand, east Cambodia, and Malaysia (2) Mekong snail-eating turtle M. subrtijuga distributed in Cambodia, southern Vietnam, southern Laos, and the lower Mun River in northeastern Thailand and Java (Indonesia); and (3) The Khorat snail-eating turtle M. khoratensis distributed in northeastern Thailand and central Laos1,2,3,4,5. The morphological characteristics, such as the pattern of the first vertebral scute, parts of lower marginal scutes 8–12, the position and shape of the infraorbital stripe, the number of nasal stripes, the number of orbital rings, and the postocular stripe, have been used to identify Malayemys species. However, these characteristics of each Malayemys species are very similar and sometimes difficult to use for species identification, whereas their DNA characteristics are quite distinct.The International Union for Conservation of Nature (IUCN) reported that M. subtrijuga is near threatened (NT) while M. macrocephala and M. khoratensis are the species of least concern (LC)6. In Thailand, only M. subtrijuga and M. macrocephala are classified as protected species by the Wild Animal Conservation and Protection Act, B.E. 2562 (2019), while M. khoratensis is not even though they are facing the same threats. Some local people in mainland Southeast Asia, such as Thailand, Cambodia and Laos, still consume meat and eggs of freshwater turtles. Thus, snail-eating turtles are commonly found in local fresh markets or fish markets in these countries3,4,5. In Thailand, they are commonly available for purchase at local markets primarily due to of their significance in religious practices among Thai people3,5. Furthermore, these turtle species are used in traditional medicine in China and Southeast Asia. The trade and consumption of freshwater turtles by humans are responsible for the decline in freshwater turtle populations and loss of biodiversity. Moreover, the turtle trade has made species identification more difficult, as the distribution of turtles is not reliable due to their transfer between locations through trade routes.Thailand government officials and some biologists have used the morphological characteristics of snail-eating turtles to identify turtle species (as shown in Table 1) in the field. However, we found that this method relies on human judgment and may not differentiate turtle species because of potential biases toward specific species. Consequently, a multiview analysis approach has been proposed to enhance decision reliability and accuracy by integrating weighted contributions from various perspectives obtained through deep learning approaches7,8,9,10. Recent advancements in the field have focused on enhancing the efficiency and accuracy of deep learning models in environmental applications. Techniques such as transfer learning, where a model developed for one task is adapted for another related task, have proven effective in scenarios where data are scarce. Additionally, efforts have been directed toward developing models that are resilient to variations in image quality and environmental conditions11,12.YOLOv8 Object detection13,14, a pivotal advancement within the domain of computer vision and deep learning, significantly enhances the image analysis capabilities of models. Recent developments in segmentation algorithms have contributed to this progress15,16,17. These models are extensively applied in critical fields such as satellite imagery18, agriculture19, facial imaging20, and medical imaging21, as well as the biological scientific domain22,23. In24,25,26,27,28,29 refers the turtle domain, However, this research has not been the turtle can object detection in single characteristic and specific sea turtle. Despite the ability of object detection techniques to provide precise details about the locations of objects and effectively manage occlusion, achieving real-time detection with both high speed and reliable accuracy remains a substantial challenge.Therefore, this study focuses on the development of a predictive model for identifying turtle species in the genus Malayemys from images and explores the effectiveness of various perspectives in species identification when expert analysis is unavailable11,30. The study employs a range of deep learning techniques for automated feature extraction and classification, aiming to identify which perspectives or combinations thereof are most informative for species recognition. We anticipate that certain perspectives will consistently yield high accuracy across diverse algorithms. This expectation extends to the use of YOLOv8 improved by CNNs trained specifically using images from the genus Malayemys. We predict improvements in accuracy. However, we also hypothesize that the comparative significance of different perspectives will remain constant across methods, assuming that no overfitting occurs with highly specialized CNNs. This study makes four key contributions to the literature:(1) Analysis of the information from four methods for identifying Malayemys turtle species.(2) Development of a combined method for accurate species identification.(3) Evaluation and comparison of how the accuracy varies for each view with the proposed model.(4) Automation and transfer of the best model to a web application.The organization of the paper is as follows: “Materials and methods” provides an overview of the material and methods, and “Proposed model” outlines the proposed model framework. “Experimental results” presents the experimental results for evaluating the models and a discussion of their implementation. “Conclusions” presents the conclusions of this study. We believe that artificial intelligence (AI) is a tool for freshwater turtle conservation because it is noninvasive, fast and easy to use in field identification.
