The full text of this item is not available at this time because the student has placed this item under an embargo for a period of time. The Libraries are not authorized to provide a copy of this work during the embargo period, even for Texas A&M users with NetID.
Artificial Intelligence-Based Computer Vision for Rapid Detection and Classification of Objects in Agricultural Situations
Abstract
In the past few years, computer vision (CV) has made significant progress due to successive improvements in computing hardware alongside machine learning (ML) and deep learning (DL) algorithms. The progressive improvements in ML and DL algorithms have enabled artificial intelligence (AI)-based CV applications. However, the agriculture sector has lagged in harnessing the immense potential of AI-based CV. Therefore, through this thesis work, applications of different AI-based CV algorithms in solving three major challenges faced by the U.S. cotton industry (detecting volunteer cotton (VC) plants in corn fields for spot-spray applications, detecting plastic shopping bags in cotton fields and classifying images of cotton leaf grades) have been presented.
VC plants growing in the middle of inter-seasonal crops like corn and sorghum can act as hosts for boll weevil pests and therefore they need to be detected, located, and sprayed. Both YOLOv3 and YOLOv5 were used to detect and locate VC plants in corn fields at three different growth stages (V3, V6 and VT) using aerial remote sensing imagery. Both the algorithms were able to detect the VC plants at accuracies greater than 90%; however, due to faster inference speed, YOLOv5 is recommended during the V3 growth stage of corn plants for near to real-time detection. The GPS coordinates of detected VC plants were used to generate optimal flight path using ant colony optimization algorithm for spot-spray applications with unmanned aircraft systems.
Plastic contamination in cotton is a serious and prevalent issue that incurs an annual loss of more than 750 million USD to the U.S. cotton industry. One of the many sources of plastic contaminants is plastic shopping bags getting carried away by wind and then tangling on cotton plants. These bags get mixed with cotton fibers during harvest and pose problems at cotton gins beside contaminating the fibers and reducing its grade. Therefore, they need to be detected and located before harvest to minimize the amount of contamination that may end up at cotton gins. It was found that YOLOv5 could detect white and brown color bags with accuracies of 92% and 85% respectively. YOLOv5m was found to be the most desirable variant among the four (others being YOLOv5s, YOLOv5l and YOLOv5x) keeping the mAP above 90% and average inference speed of about 86 frames per second.
In the final study of this thesis, a custom VGG16 network was used with softmax and support vector machine (SVM) classifier for classifying images belonging to five cotton leaf grades (i.e., grades 2, 3, 4, 5 and 6). It was found that SVM in the custom VGG16 network could achieve the same accuracy as softmax classifier but at a much smaller computation time.
Subject
computer visionartificial intelligence
volunteer cotton
boll weevil
YOLOv3
YOLOv5
VGG16
unmanned aircraft systems
ant colony optimization, spot-spray, plastic contaminants
cotton leaf grade
support vector machine
Citation
Yadav, Pappu Kumar (2022). Artificial Intelligence-Based Computer Vision for Rapid Detection and Classification of Objects in Agricultural Situations. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /198137.