Enhanced Threshold-based Segmentation for Maize Plantation

Abstract

Phenotyping, mainly plant’ health monitoring, is labor-and time-intensive, particularly for large-scale operations like maize plantations. Therefore, this research used a drone equipped with an RGB image to photograph the whole plantation quickly. On the other hand, RGB photographs do not categorize plants and weeds due to high brightness, shadows, and overlapped foliage. Therefore, several segmentation algorithms are used to solve various challenges. For instance, threshold-based segmentation can only accept progressive illumination, which is crucial for outdoor lighting, simplicity, and distinguishing objects with identical hues. For this kind of segmentation, however, intense light requires modification. Consequently, threshold-based segmentation was improved to normalize the disturbances above while rapidly separating leaves from weeds. In this manner, the Enhanced threshold-based segmentation had applied to RGB images of maize plantations like cornfields with distractions seen in the gathered photos with a segmentation accuracy of 92.41%. In comparison, the threshold-based segmentation had used in the same dataset without normalizing the picture's luminance, with a segmentation accuracy of 5.71%. Thus, the enhanced segmentation method improved segmentation accuracy by 86.7% compared to threshold-based segmentation, which is limited to extreme light conditions. Thus, the incorporated normalization in the segmentation process significantly increases the segmentation accuracy.