Date of Award
5-2025
Degree Type
Thesis
Degree Name
M.S.
Degree Program
Computer Science
Department
Computer Science
Major Professor
Tamjidul Hoque, PhD
Second Advisor
Christopher Summa, PhD
Third Advisor
Shreya Banerjee, PhD
Abstract
Animal burrow detection is a time-consuming and costly task for levee inspectors. Annual budgets run up to approximately $16 million per state. The inspectors typically will have to travel to the inspection sites using government-assigned transportation. Depending on the distance to the site, it may take minutes or hours to arrive before any productive inspections occur. Once at the site, the inspectors were subject to human error, overgrown foliage, severe weather, or prohibitive landscaping that would make any human inspection impossible. Also, animal burrows could be small enough or overgrown, so the human inspector misses the problem areas. We aimed to aid the inspector with a simplified method of inspecting levee systems. We trained ResNet-34 on animal burrow images taken by manual inspection. The model was trained on both positive (animal burrow) and negative (no animal burrow) images using many iterations of the model. Iterations included changing the model weights to classify burrows and bounding boxes where the burrows were located to determine the best accuracy. Batch normalization was used to improve the speed and stability of training the network by normalizing the inputs using re-scaling and re-centering. The initial convolutional layer kernel size was changed from 7 × 7 to 3 × 3. This is because it allows the capture of local and global features of the images. An initial learning rate of 0.0005 was used to train the model, but eventually, this was switched to an exponential learning rate to mitigate noise in the training accuracy. The testing showed that classifying an image as positive or negative with ResNet-34 was effective for extensive burrows, but accuracy suffered with small burrows.
Recommended Citation
Moore, Christopher D., "Animal Burrow Detection in Levee Systems Using Res-Net 34" (2025). University of New Orleans Theses and Dissertations. 3285.
https://scholarworks.uno.edu/td/3285
Rights
The University of New Orleans and its agents retain the non-exclusive license to archive and make accessible this dissertation or thesis in whole or in part in all forms of media, now or hereafter known. The author retains all other ownership rights to the copyright of the thesis or dissertation.