TALOS: optimization of the CNN for the detection of the tomato leaf diseases

Abstract

Early detection of plant diseases using convolutional neural network (CNN)is crucial for maximizing crop yield and minimizing economic losses. Manual inspection, the frequent technique, is inefficient and error prone. While CNN’s offer potential for accurate and quick disease recognition, their performance is highly dependent on effective hyperparameter tuning. This process is time consuming, resource intensive, and needs significant expertise due to the vast hyperparameter space, since it can be hard to figure out which is ideal for optimal performance. An effective optimization tool, tunable automated hyperparameter learning optimization system (TALOS), is proposed, which automates the tuning of hyperparameters by systematically exploring the hyperparameter space and evaluates different combinations of parameters to find the optimal configuration that maximize the model’s performance. The performance of this approach is recognizable through its exploration of five different hyperparameters across a search space of 32 combinations, yielding optimal parameters by the second round. Using 3030 tomato leaf images from a benchmark data set, the model achieves a remarkable 94.7% validation accuracy with 33647 trainable parameters. Thus, automated hyperparameter tuning approach not only optimizes model performance but also reduces manual effort and resource requirements, paving the way for more effective and scalable solutions in agricultural technology.