The proposed deep learning framework addresses the challenge of efficiently processing and analyzing aerial imagery to detect drought stress in potato crops. It utilizes a transfer learning approach that combines a pre-trained convolutional neural network (CNN) with custom layers for targeted dimensionality reduction and enhanced regularization. This architecture effectively leverages the feature extraction capabilities of the pre-trained network while the custom layers enable improved performance on the specific task of drought stress identification.
A key innovation of this work is the integration of Gradient-Class Activation Mapping (Grad-CAM), an explainability technique that sheds light on the internal workings of the deep learning model. Grad-CAM visualizes the regions within the input images that the model focuses on to make its predictions, fostering interpretability and building trust in the model's decision-making process.
The framework was evaluated using a dataset of aerial images of potato crops, and the results demonstrate its superior performance compared to existing state-of-the-art object detection algorithms. The proposed pipeline with the DenseNet121 pre-trained network achieved a precision of 98% for the stressed class and an overall accuracy of 90%. The explainable nature of the model, combined with its high accuracy, makes it a powerful tool for drought stress identification in potato crops, enabling informed decision-making and timely intervention.
To Another Language
from source content
arxiv.org
Önemli Bilgiler Şuradan Elde Edildi
by Aswini Kumar... : arxiv.org 04-17-2024
https://arxiv.org/pdf/2404.10073.pdfDaha Derin Sorular