Towards Automatic Detection of Garlic Insect Pests and Diseases in Digital Images

Pest and disease detection on crops have been well investigated by the research community in previous years. Most of the proposed methods make use of sophisticated deep-learning models that would require higher resource requirements. Recent works in this area suggest that researchers are moving towards resource-friendly alternatives. Also, detecting insect pests and diseases in garlic crops is a very challenging task as indicators happen at the early stages of plant growth. This would mean that most of the field environments have the same color profile which makes the recognition process more difficult. In this research, a recognizer with 4 convolutional layers is proposed to detect insect pests and diseases in the garlic crops. The data was gathered from the field during the course of 1 year from October 2021 to November 2022. There are 444 garlic samples which were classified into 7 groups namely, Common Cutworm, Leaf Spot, Purple Blotch, Common Worm, Tangle Top, Root Rot, Garlic Rust, and Healthy, based on the perspective of 3 garlic experts. The proposed model was trained using 80% of the dataset while the other 20% was used for testing for a period of 9,000 epochs. The recognizer was able to manage a 97% accuracy score while maintaining a Sensitivity of 86% for all the test cases. The Matthews Correlation Coefficient of the proposed model is 0.45 which suggests that the model is predictive without sacrificing more resource requirements.