Student Research Reports
Open social media sharing options menu
A Comparative Analysis of Different Convolutional Neural Network Architectures for Mosquito Genera Classification
Country:United States of America
Student(s):Dhilan Shah, Suhani Shukla, Michael Squeri, Sameen Ahmad, Ankhi Banerjee, Adriana Talianova, Saurav Bavdekar, Arav Sachdeva, and Arnav Deol
Grade Level:Secondary School (grades 9-12, ages 14-18)
GLOBE Teacher:Cassie Soeffing
Contributors:Dr. Rusty Low, IGES, scientist
Peder Nelson, OSU, sme
Dr. Erika Podest, NASA JPL, scientist
Andrew Clark, IGES, EO Researcher and Data Analyst
Report Type(s):International Virtual Science Symposium Report, Mission Mosquito Report
Protocols:Earth As a System, Mosquitoes
Presentation Video:
View Video
Presentation Poster:
View Document
Optional Badges:I am a Data Scientist, I make an Impact, I work with a STEM Professional
Language(s):English
Date Submitted:01/24/2023
Vector-borne diseases, such as Dengue Virus, Zika Virus, Malaria, and West Nile virus, cause an
estimated 700,000 annual deaths. Aedes, Culex, and Anopheles, three common mosquito genera, carry and transmit these diseases. Climate change is worsening the frequency and severity of infection, and currently, there is a limited number of universally accessible cures. Mosquito genera’s disease-carrying capacities vary due to their unique adaptations, making classification valuable in understanding the susceptibility of some viruses. Additionally, identifying mosquito genera can inform preventative measures to reduce disease transmission and help health professionals determine proper mitigation measures. With this information in mind, our group built a research project centered around the following question: which convolutional neural network (CNN) architecture can most effectively distinguish between Aedes, Anopheles, and Culex mosquito larvae? First, we extracted data from the GLOBE API. Then, using a Python algorithm and metadata containing mosquito genera, we sorted the images by genus. Finally, we trained four CNNs and compared their ability to identify mosquito genera with image classification. The architectures used for larvae classification were LeNet-5, AlexNet, VGG-16 Net, and ResNet-50. Each model was built in Google Colab in Python using Tensorflow and Keras libraries, trained using the larvae images, and analyzed to identify the best overall network. After the trials, ResNet-50 was the most effective model due to its low loss of 0.60. Understanding the accuracy of CNN architectures when identifying mosquito genera can support further research in the scientific community regarding computer vision techniques in biological fields. Most importantly, combining this research with other mosquito tracking algorithms can save lives, specifically in areas most sensitive to mosquito-vectored diseases.
Key Words: Convolutional Neural Networks, Mosquito Larvae, Image Classification, Artificial Intelligence, Data Science