SAM-IE: SAM-enabled image enhancement for segmentation of infected cucumber leaves

Abstract

The synthesis of compounds in plants with the aid of radiant energy is essential for plant growth and crop yield. However, the leaves through which this synthesis of compounds occurs, when infected, cannot perform this function optimally and thus negatively impact plant growth and crop yield. Symptoms of cucumber disease are observed on the cucumber leaves. The main objective of this paper is to facilitate automated detection systems for the segmentation of infected cucumber leaves from healthy leaves by proposing a novel SAM-IE (Segment Anything Model-Image Enhancement) model using ResNet50 and CNNs. The SAM-IE model is a hybrid of an existing SAM model and a novel IE method. All the data collection and processing steps were explored. Moreover, experiments were performed, and competitive results were obtained with detailed analysis as follows: ResNet50 with SAM-IE obtained 0.885 accuracy, 0.894 F1-Score, and a processing time of 112 (ms). ResNet50 without SAM-IE obtained 0.883 accuracy, 0.845 F1-Score, and a processing time of 121 (ms). CNNs with SAM-IE obtained 0.872 accuracy, 0.831 F1-Score, and a processing time of 115 (ms). CNNs without SAM-IE obtained 0.772 accuracy, 0.821 F1-Score, and a processing time of 123 (ms). To further validate these results, we compared them with the existing results from K-means clustering, Fuzzy C-means clustering, Expectation Maximization (EM), and Superpixels + EM. The practical implications of the findings in this paper are essential for the horticulture industry. Horticulture farming systems that incorporate deep learning not only enhance plant growth but also ensure high crop yield via automated management and monitoring, reducing crop vulnerability to disease and increasing economic gain.