Plant Methods, Год журнала: 2025, Номер 21(1)

Опубликована: Май 30, 2025

Rice is one of the major food crops in world, and monitoring its growth condition great significance for guaranteeing security promoting sustainable agricultural development. Leaf area index (LAI) a key indicator assessing yield potential rice, traditional methods obtaining LAI have problems such as low efficiency large error. With development remote sensing technology, unmanned aerial multispectral combined with deep learning technology provides new way efficient accurate estimation rice. In this study, camera mounted on UAV was utilized to acquire rice canopy image data, uniformly estimated over multiple periods by multilayer perceptron (MLP) convolutional neural network (CNN) models learning. The results showed that CNN model based five-band reflectance images (490, 550, 670, 720, 850 nm) input after feature screening exhibited high accuracy at different stages. Compared MLP vegetation indices inputs, could better process original effectively avoiding problem saturation, improving accuracies 4.89, 5.76, 10.96, 1.84 6.01% tillering, jointing, booting, heading periods, respectively, overall improved 6.01%. Moreover, (MLP CNN) before variable noticeable changes. Conducting contributed substantial improvement estimation. an method unified multi-period LAI. generalization ability adaptability were further rational data enhancement techniques. This study can provide technical support precision agriculture more solution monitoring. More extraction be explored future studies optimizing structure improve stability model.

Язык: Английский