Abstract:
One of the top machine learning algorithms for image classification is Convolutional Neural Networks, according to experts. Convolutional Neural Networks has been extensively used in the agriculture sector for a variety of solutions, including identifying plant diseases, forecasting crop production, and categorizing land cover, among others. Unfortunately, creating Convolutional Neural Networks models requires a significant amount of data, which is extremely difficult to come by in agriculture. In this study, we suggest combining the Convolutional Neural Networks and Multitask Learning techniques since it has been shown to be a good algorithm to use when there isn't enough data by utilizing its ability to share layers. Additionally, Multitask Learning enabled us to simultaneously identify pathogens and diseases that affect maize, which is not possible when using a single Convolutional Neural Networks model. Indeed, recognizing pathogen may help at preventing the disease to spread throughout the whole field. Multitask Learning helped in improving the performance of our model by reducing overfitting. In this research, we combined Multitask Learning with other regularization techniques for a better performance. Indeed, the test accuracy of the overfitting model increases from 60.08% for the single maize disease model to 74.48% when combining the maize disease identification model with the maize pathogen identification model in one model using Multitask Learning. The accuracy rises to 77.44% while combining Multitask Learning to the early stopping method. However, the test accuracy goes up to 85.22 % when Multitask Learning is combined with Early Stopping and Transfer Learning.
Keywords: Multitask learning, Convolutional Neural Networks, Overfiting, Image Classification, and Regularization Methods.
