The College of Engineering Summer Undergraduate Research Program (SURP) is a unique opportunity for college of engineering undergraduate students to engage in hands-on research with a faculty mentor while using critical thinking, collaborative, and entrepreneurial skills. SURP is an 8-week long program (can be non-contiguous), and students are expected to spend at least 20 hours/week working on their research projects and additional time preparing and presenting their poster for the annual SURP symposium; SURP students and their faculty mentor will agree upon a start date and timeline for their research project (with a suggested start date of June 26th, 2023). The goals of SURP are to: 

Project Title: Deep-Hash Vision: Drone based Deep Hashing for Livestock Monitoring

Faculty Advisor: Sumona Mukhopadhyay

Faculty Email: mukhopad@calpoly.edu

Description of Research Project: This project focusses on developing a Machine Learning (ML) / Deep Learning (DL) system for monitoring of livestock using aerial images. Building on the work done initially as a SURP 2022 project, the focus for this project this year will be to improve and accelerate the detection of several different livestock in the images using ML/DL . During SURP 2022, a benchmark model was developed for livestock detection of a single animal type using aerial images. The detection was found to be slow and worked for a single livestock detection. This year we focus on improving the model by using semantic hashing technique which are unique codes for an image that could accelerate the detection. In order to speed up the monitoring task using semantically similar livestock images, which is a crucial requirement for real-time detection, we propose to represent the images as a binary code or the hash code. Hashing has been a widely-adopted technique for nearest neighbor search which is a machine learning approach for image recognition in large-scale image retrieval tasks. This increases the recognition rate and allows for herd or group recognition of livestock. The livestock are detected based on using the hash codes with the popular nearest neighbor detector. Moreover, semantic based hashing will allow to retrieve images determine the highest possible accuracy that could be achieved in the detection of animals, of livestock which are visually similar to other breeds. Moreover, we want the model to be scalable such that it detects multiple different categories of livestock in the images. Thus an end-to-end ML pipeline for realtime detection of different types of livestock are the ultimate goal of this project.

HOW TO APPLY: