Abstract
Deep learning was used in this study to recognize endangered flower species in Kazakhstan. The results can help monitor biodiversity and protect those species. Convolutional Neural Network (CNN) models were used to classify different flower species. In this project, a publicly available Kaggle dataset was used, that included five flower species, namely sunflower, tulip, dandelion, rose, and daisy and two endangered species from the Red Book of Kazakhstan, Paeonia anomala and Crocus korolkowii. Employing transfer learning with pre-trained ImageNet weights, eight deep learning architectures were trained and evaluated: AlexNet, VGG16, GoogLeNet, ResNet50, DenseNet121, EfficientNet-B0, MobileNetV2, and YOLOv8-cls. To enhance generalization and address class imbalance, data augmentation and class equalization methods were applied. The evaluation criteria included accuracy, precision, recall, F1-score, and mean Average Precision (mAP). When comparing the models, YOLOv8-cls yielded the highest accuracy, whereas MobileNetV2 provided a noteworthy balance between speed and precision. This research reveals that image recognition systems rooted in deep learning can benefit ecological studies conducted in Kazakhstan by automatically recognizing common and endangered flowers, therefore playing a role in cataloging biodiversity and environmental monitoring.