Font Size: a A A

Butterfly Target Detection Based On MobileNet

Posted on:2021-01-10Degree:MasterType:Thesis
Country:ChinaCandidate:Y YangFull Text:PDF
GTID:2480306341984189Subject:Master of Agriculture
Abstract/Summary:PDF Full Text Request
Butterflies are an important link in the insect biological chain,and related scientific research is of great significance to the protection of the ecological environment in China.Although scholars have made many achievements in insect recognition,related research on butterfly recognition is still lacking,and how to improve image recognition speed and recognition accuracy is still an important issue that needs to be resolved.Therefore,based on the butterfly image data set,MobileNet-SSD,Faster R-CNN,and YOLO v3 were constructed to identify and detect butterfly insects,and four performance indicators:Precision,Recall,Mean Average Precision(m AP),and Frames Per Second(fps)were used to verify the effect of the models.This research aims to obtain a method for identify butterfly insects more quickly and accurately,and to provide reference ideas for butterfly insect identification and detection.The main conclusions of this study are as follows:(1)This paper used web crawling combined with the number of existing butterflies to construct the butterfly insect target recognition data set required for the research,it contains a total of 16 butterfly species and a total of 1615 pictures..the original data were expanded by using image data augmentation(rotation,flip,zoom,shift,etc.).After data enhancement,it was expanded to 17622(2)The three models used in this study were optimized by adjusting the learning rate,selecting the optimization algorithm,setting the training batch size value,and the number of iterations to obtain the optimal index.(3)The MobileNet-SSD target detection algorithm performs best in this research.The average accuracy(m AP)value on the test set is 99.63%,which is 9.65% higher than Faster R-CNN;the average number of detections per second is 16.30 fps,which is 6.80 fps higher than Faster R-CNN and higher than YOLO v3 5.78 fps.This method can effectively identify butterfly insects more quickly and accurately.(4)Through Web App technology,the training model is implanted in the mobile terminal,and the mobile application of identifying butterfly insects is developed.
Keywords/Search Tags:butterfly recognition, MobileNet, convolutional neural network, target recognition algorithm, image recognition
PDF Full Text Request
Related items