Mahalakshmi S, Jose Anand A, Partheeban Pachaivannan
Department of Computer Science and Engineering, Chennai Institute of Technology, Chennai, Tamil Nadu, India.
Department of Electronics and Communication Engineering, KCG College of Technology, Chennai, Tamil Nadu, India.
J Environ Manage. 2025 Apr;380:125095. doi: 10.1016/j.jenvman.2025.125095. Epub 2025 Mar 25.
Crop yield is a significant factor in world income and poverty alleviation as well as food production through agriculture. Conventional crop yield forecasting approaches that employ subjective estimates including farmers' perceptions are imprecise and contain high variability over large farming areas, particularly in areas where data is limited. The improvement of data capture techniques in the last few years especially from high-resolution sensors and Deep Learning (DL) have enhanced the quality and scope of agricultural data to assist policymakers and administrators. Mostly researchers used various techniques for independently forecasting soil fertility and crop yield. In image processing, Sentinel-2 is one technique that enhances agriculture, especially in analyzing crop health and type of soil prediction. Using the Normalized Difference Vegetation Index (NDVI) for processing the red and near-infrared bands allows computation ranges between -1 and 1. The values are higher than 0.7, the crops are in good health, or the values are less than 0.3 means crops are under stress. Therefore, information about soil types and NDVI data provide the most elaborate recommendations regarding agriculture. This is done through executing superior picture analysis and verification for precise errors below 5 %. It also develops a rainfall-runoff forecast through a Convolutional Neural Network approach. Our proposed methodology attains an average accuracy of about 98.7 % compared with traditional approaches average is about 85 %-90 %. A high-accuracy model of this type facilitates a spatial and temporal resolution of five days and improves farmers' irrigation process since it offers more accurate agronomic decisions. This research may lead in the agriculture and deep learning applications for economic and societal improvement. Application of artificial intelligence in agriculture synchronizes relevancy from satellite imagery making precision smart and boosting food productivity by 20 % with better utilization of resources.
作物产量是世界收入、减贫以及农业粮食生产的一个重要因素。采用包括农民认知在内的主观估计的传统作物产量预测方法并不精确，且在大面积种植区域存在很大差异，尤其是在数据有限的地区。过去几年数据采集技术的改进，特别是来自高分辨率传感器和深度学习（DL）的技术，提高了农业数据的质量和范围，以协助政策制定者和管理人员。大多数研究人员使用各种技术独立预测土壤肥力和作物产量。在图像处理中，哨兵 - 2 是一种增强农业的技术，特别是在分析作物健康状况和土壤类型预测方面。使用归一化植被指数（NDVI）处理红色和近红外波段可使计算范围在 -1 到 1 之间。值高于 0.7 表示作物健康状况良好，值小于 0.3 则意味着作物处于压力之下。因此，关于土壤类型和 NDVI 数据的信息提供了有关农业的最详尽建议。这是通过执行高级图像分析和验证来实现的，以确保精确误差低于 5%。它还通过卷积神经网络方法开发降雨径流预测。与传统方法平均约 85% - 90% 的准确率相比，我们提出的方法平均准确率约为 98.7%。这种高精度模型有助于实现五天的空间和时间分辨率，并改善农民的灌溉过程，因为它提供了更准确的农艺决策。这项研究可能会引领农业和深度学习在经济和社会改善方面的应用。人工智能在农业中的应用使卫星图像的相关性同步，实现精准智能，并通过更好地利用资源将粮食生产率提高 20%。
