Maturity Grading and Macronutrient Deficiency Identification to Assist Tomato Growth in a Wick Hydroponic System Using an Xception-Based Convolutional Neural Network

Abstract

Tomatoes are extensively cultivated around the world and are considered to be one of the most lucrative crops in the Philippines. Growing tomatoes is considered as a high-risk activity since it is susceptible to disease and pest attacks, and fluctuations in nutrient intake. One solution to these issues is to cultivate them in protected environments such as hydroponic systems. With these types of systems, it is important to have efficient classification of tomato maturity and accurate assessment of nutrient deficiencies to result in optimal growth and yield since conventional methods are time-consuming and prone to error. Several studies have been conducted to address these issues using image processing and deep learning. In this study, two Xception-based Convolutional Neural Network models were developed for tomato maturity grading and macronutrient deficiency identification in tomato plant leaves grown in a Wick Hydroponic System in the prospect of presenting better results among other existing approaches in classifying tomato maturity stages and macronutrient deficiencies. After gathering tomato and leaf images using a smartphone camera, a pre-trained image classification Xception model was retrained and optimized to classify unripe, half-ripe, and ripe tomatoes for maturity grading, and healthy, nitrogen-deficient, phosphorus-deficient, and potassium-deficient tomato plant leaves for macronutrient deficiency identification. The initial models indicated signs of overfitting but after adding Ll regularization and batch normalization, and reducing the number of dense layers, the maturity grading and macronutrient deficiency identification models obtained an accuracy of 93.3% and 97.3% respectively.