자료유형  

 학위논문

Control Number  

0016935453

International Standard Book Number  

9798380182461

Dewey Decimal Classification Number  

630

Main Entry-Personal Name  

Wiegman, Christopher R.

Publication, Distribution, etc. (Imprint  

[S.l.] : The Ohio State University., 2021

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2021

Physical Description  

1 online resource(142 p.)

General Note  

Source: Dissertations Abstracts International, Volume: 85-03, Section: B.

General Note  

Advisor: Shearer, Scott A.

Dissertation Note  

Thesis (Ph.D.)--The Ohio State University, 2021.

Restrictions on Access Note  

This item must not be sold to any third party vendors.

Summary, Etc.  

요약Crop stress continues to cost American farms significant portions of yield. On average, annual yield loss to disease in North America are estimated at 11% for soybeans and anywhere from 2% to 17% for corn. Although critical, modern scouting practices still largely rely upon manual scouting and can fail to detect stresses before yield loss occurs. Technological solutions represent one option to address this need in farming communities, not only to protect yield, but help reduce agriculture's environmental footprint. Despite growth the field of precision agriculture with respects to precision application, there are still significant challenges surrounding automated crop stress classification due to the complexity of the task.Modern computational tools in the field of computer vision like Convolutional Neural Nets (CNN) offer significant promise in their ability to classify crop stress using basic red-green-blue imagery. These algorithms have demonstrated significant performance in a variety of applications ranging from industrial to medical and even military; however, stress classification is pushing these algorithms to their performance limits. Agriculture provides a unique set of challenges such as deployment scale, high spatial resolution requirements, limited access to training data, high levels of both intraclass variation and inter-class similarities along with a low error tolerance. This work details one approach, starting with developing an intra-canopy sensor platform for improved data acquisition called the 'Stinger', conducting algorithm training with an in-depth study of data augmentation, and finally, back-end modification of the classification output with a hierarchy, to provide potential solutions to these challenges and mitigate their impact on classifier performance.The Stinger demonstrated an ability to acquire intra-canopy imagery, the ideal data for crop stress classification with CNNs, with the speed of a small Unmanned Aerial System (sUAS) thus alleviating current limitations with respect to manual CNN deployment on mobile smartphones. Results from comprehensive experimentation with image augmentation for classifier training indicated reliable gains in accuracy between 5 and 10% when multiple geometric and photometric augmentations were used compared to a control classifier with an accuracy of 51%. These results provide a concrete set of best data augmentation practices to help ensure application specific classifiers can generalize and maintain performance upon deployment to unseen data outside of testing. Finally, work with confidence driven hierarchical classification demonstrated significant improvements in classification performance. A final algorithm accuracy of over 93% was attained with a base classifier capable of only 68% accuracy representing a gain in accuracy of 25%. Not only does this demonstrate a significant improvement in performance but details the limits of the hierarchical approach to compensate for an underperforming base classifier.The guiding principle of this work was to experiment with novel approaches such as intra-canopy sensing and confidence driven hierarchical classification to help make automated stress surveillance with CNNs a reality for producers. This goal was largely accomplished as the results presented demonstrate significant improvement compared to the current state of the art.

Subject Added Entry-Topical Term  

Agriculture.

Subject Added Entry-Topical Term  

Agronomy.

Subject Added Entry-Topical Term  

Food science.

Subject Added Entry-Topical Term  

Plant pathology.

Index Term-Uncontrolled  

Crop stress

Index Term-Uncontrolled  

North America

Index Term-Uncontrolled  

Soybeans

Index Term-Uncontrolled  

Convolutional Neural Nets

Index Term-Uncontrolled  

Small Unmanned Aerial System

Added Entry-Corporate Name  

The Ohio State University Food Agricultural and Biological Engineering

Host Item Entry  

Dissertations Abstracts International. 85-03B.

Host Item Entry  

Dissertation Abstract International

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:642900