Research On Butterfly Species Recognition Based On Deep Learning

With the continuous development of society,people are increasingly aware of the importance of protecting animals and maintaining biodiversity.The need for accurate and consistent identification of various species is also growing.The species of insects is the most abundant among all types of animals.As one of the most abundant insects,there are nearly 20,000 butterflies in the world.Butterflies have complex colors,textures,and patterns,which makes manual recognition difficult and costs lots of time.Therefore,research on the automatic identification of butterfly species has received increasing attention.The rapid development of computer hardware and software has led to a rapid increase in its ability to process image data.Based on this,related technologies of image recognition have also continuously made new breakthroughs.Deep learning has developed rapidly in recent years and has excellent performance on computer vision tasks.In order to extracting feature maps from original image,it combines low-level features into more abstract high-level representations by a series of non-linear mappings.Then,the extracted features are feed into classifier to achieve the purpose of recognition.Through the backpropagation algorithm,it can automatically learn network parameters to extract effective features,thus the traditional methods of manually extracting features were abandoned.Therefore,deep learning can be used to analyze large-scale data.So we apply the deep learning to the automatic identification task of butterfly species,which can automatically locate the position of the butterflies in the image,and extract their features for classify or clustering,and improve the efficiency of butterfly species identification.In the past,the automatic identification researches of butterflies were mostly based on butterfly specimen images dataset.The number of butterflies and the number of images involved in the dataset were small,and most of the research use artificial design methods to extract effective features.Automatic identification of butterfly species based on butterfly photos taken in the ecological environment is a challenge to the existing butterfly species identification,because the classification features of butterflies in the ecological environment will be severely obscured.Therefore,this study is based on the deep learning method to research the automatic identification of butterfly species in ecological photos,including the location and species identification of butterflies in ecological photos.It is very difficult to label the butterfly species in the photos of butterflies in the ecological environment,which requires professionals to identify butterfly species,even challenging for professionals.Therefore,by virtue of the strong learning ability of deep learning on features,this study realizes the identification of butterfly species by supervision and semi-supervision methods.The main work of this study is as follows:1.Using the Faster R-CNN and Mask-RCNN object detection algorithm based on deep learning to locate the positions and identify the species of butterflies in the ecological picture.The butterflies training set is constructed by adding corresponding butterfly specimen images,and the training set is expanded by rotating?adding noises.The problem of small-scale data,and fine-grained classification of small species under the same large species.We use detection evaluation criteria mAP to view the effects of butterfly positioning and recognition.The experimental results show that the Mask-RCNN algorithm can achieve better effects in the automatic identification of butterfly species.2.Due to the difficulty of labeling butterfly datasets,it is necessary for professionals to label species and take a long time.This study explores the problem of automatic identification of butterfly species on ecological images when the artificially labeled butterfly data is insufficient.We use the object detection algorithm to detect the position of the butterfly in the ecological photos,and DeepCluster algorithm to partition the detected butterfly according to the species.Then the butterfly species are identified semi-supervised.