Computer Vision & Image Processing II

C3L-C: Computer Vision & Image Processing II

Session Type: Lecture
Session Code: C3L-C
Location: Room 3
Date & Time: Friday March 24, 2023 (11:20-12:20)
Chair: Paris Giampouras
Track: 5

Paper ID	Paper Title	Authors	Abstract
3052	A Deep Transfer Learning Based Approach for Forecasting Spatio-Temporal Features to Maximize Yield in Cotton Crops	Krishna Chaitanya Gadepally{2}, Sambandh Bhusan Dhal{2}, Mahendra Bhandari{1}, Juan Landivar{1}, Stavros Kalafatis{2}, Kevin Nowka{2}	Cotton is an important economic crop farmed in the United States. Monitoring cotton crop growth metrics during in-season growth, from early season growth to harvest, is critical. Because cotton crop output is directly related to management decisions made to regulate growth parameters during a cultivation season, utilizing forecasting models to predict future values of canopy indices has piqued the interest of researchers. In this paper, we have used the canopy feature data i.e., Canopy Cover, Canopy Height and Excess Green Index recorded in the year 2020 and trained a multi-layer stacked LSTM model. Next, a Deep Transfer Learning based approach was used to freeze the weights of the initial layers of the trained LSTM model, and the weights of the last few layers were fine-tuned based on the 2021 cultivation year canopy index data to predict the canopy features from 28th day of cultivation to the end of the harvesting period.
3081	Tools and Visualizations for Exploring Classification Landscapes	William Powers, Lin Shi, Larry Liebovitch	Neural networks and deep learning systems find the correct classification of input data by locating the corresponding local minima in the hyper-dimensional, classification landscape. An increasing number of adversarial examples have now shown that these networks sometimes find an unexpected and incorrect minimum and so make an incorrect classification. To understand those results requires a better understanding of the nature of these classification landscapes. Previous studies have explored the properties of the landscape of back propagation in training these networks. In our studies here, we explore the classification landscape of already trained networks. We present some novel procedures and analytical tools to study the classification landscape and visualizations to meaningfully represent those results. We apply these methods to study the classification landscape in classic examples, including image classification in the MNIST data set and flower classification from numerical feature values in the Iris data set.