Deep Learning for Assessing Unhealthy Lettuce Hydroponic Using Convolutional Neural Network based on Faster R-CNN with Inception V2

Abstract

The hydroponic system is a development of traditional farming that substitute soil as a medium plant due to land limitation. Lettuce is the most popular hydroponic vegetable product in the market. However, during harvesting, there are huge challenges to ensure product quality especially for mass production has a better quality. In this research, we utilized Deep Learning as objection detection to recognize the disease in Hydroponic vegetables by using Faster R-CNN with Inception V2 algorithm and compare the performance by divided the ratio of training and validation dataset into 3 categories i.e. 78/9, 70/17, and 61/26 with the standard testing ratio for all categories is 13%. From this study we obtain a result that ratio 78/9 have a better performance with Accuracy 70%; Precision 97%; Recall 68% and F1 Score 80% however, ratio 61/26 has the lowest performance with Accuracy 40%; Precision 24%; Recall 100% and F1 Score 38,5% from 412 images dataset with 53 testing images with default learning rate setting 0.0002. As the result shown that the testing and validation ratio was affected by the deep learning model performances.