Recognition of Butterflies of Bangladesh using Deep Learning

Butterflies have the capacity to detect and respond to rapid climate change as well as have a great impact in the field of entomology. There are different varieties of butterflies, and the study of butterfly categorization is important for practical activities such as ecological sustainability and pest management in agriculture and forests. The traditional butterfly identification procedure relies on morphological characteristics such as shape, color, wingvenation, and patterns, as well as dissection experiments and molecular methods, which are complex, expensive, laborious, consume a lot of time, and may require data collection, recognition, and physically preserving specimen images. However, because of their high similarity, wide variety, and distinguishing characteristics, the classification of butterflies has slow recognition speed and low accuracy at this stage. For this reason, conducting and improving the effectiveness of automatic identification of butterflies is significant. Therefore, the deep convolutional neural network-based detection model YOLOv4 is applied in this work which is an automated technique that does not require human intervention and the other three pre-trained deep learning models have been experimented with namely SSD-MobileNet v2, YOLOv5, and Faster R-CNN to compare the performance of the suggested system. In this study, a novel butterfly dataset is created with 1,640 ecological images of butterflies belonging to 15 commonly found species in Bangladesh. The experimentation results demonstrate that the automated butterfly recognition procedure based on the YOLOv4 architecture can detect and identify butterfly photographs in an environmental context automatically, with a mean average precision (mAP) value of 98.09%.