Thesis
Performance Comparisons of Existing Deep-Learning Models for Crop Disease Species Classification and Crop Disease Severity Classification
Masters by Research, Murdoch University
2024
Abstract
The Australian agricultural industry is currently facing numerous related challenges. These challenges can trace their causes to two broad domains: human labour and pests. The cost of human labour and chemical pesticide application on farms accounts for approximately 67.5% of the average Australian farm’s operational expenditure. This has put pressure on the agricultural industry as a whole, and a practical solution needs to be developed. One solution which has been proposed is the implementation of intelligent robotic labour on farms to assist human labour and replace chemical pesticides. However, these crop operation robots are not yet commercially viable due to their poor performance, particularly in regards to the accuracy and efficiency of the robot’s crop disease classification model. Improving model performance will help to increase the commercial viability of crop operation robots. Therefore, this research aims to further progress the accuracy and efficiency of crop disease identification models, particularly in regards to disease severity.
As part of this research, previous literature has been examined to determine the current scope of automatic intelligent crop disease identification capabilities. A benchmark study was conducted to determine the current standard of the AlexNet model on crop disease species classification. Subsequently, three existing image classification models (AlexNet, MobileNetv3, and MLP-Mixer) have been trained and tested to determine the current performance of performance for crop disease species classification. These same three models were also trained and tested to determine the performance of said models when applied to crop disease severity classification.
Results from the tests conducted herein demonstrate that the performance of the AlexNet and MobileNet models for both crop disease species classification and severity classification exceed that of the MLP-Mixer model. However, the MLP-Mixer model may possess some state of the art qualities, specifically with regards to efficiency, which exceeds that of the AlexNet model.
Details
- Title
- Performance Comparisons of Existing Deep-Learning Models for Crop Disease Species Classification and Crop Disease Severity Classification
- Authors/Creators
- Adam Stevenson
- Contributors
- Hai Wang (Supervisor) - Murdoch University, Centre for Water, Energy and WasteAmirmehdi Yazdani (Supervisor) - Murdoch University, Centre for Water, Energy and Waste
- Awarding Institution
- Murdoch University; Masters by Research
- Identifiers
- 991005753924807891
- Murdoch Affiliation
- School of Engineering and Energy
- Resource Type
- Thesis
- Note
- Accelerated Research Masters with Training (aRMT)
Metrics
39 File views/ downloads
69 Record Views