The recent surge in digital agriculture has generated an emerging demand for scalable, resource-efficient solutions capable of handling both close-range images of agricultural products and high-scale remote-sensing images. Deep learning models have high accuracy, but they are expensive and lack the dynamism to be deployed in cloud-based and resource-constrained environments. To mitigate this gap, this research paper recommends Dynamic Resource Flow Control (DRFC), an efficient cloud-native feature-reduction and clustering algorithm designed to handle heterogeneous agricultural imagery and minimize the number of computational tasks assigned to distributed nodes. DRFC merges lightweight dimensionality reduction with active resource flow management and dynamically allocates cloud resources, maintaining the discriminative nature of the high-dimensional data structure. The framework has been tested on two benchmark datasets: Fruits-360 for product-level classification and the USDA Cropland Data Layer/BigEarthNet for crop-level analysis at the remote sensing scale. Measures of performance include accuracy, F1-score, mAP, and resource-efficiency measures, and DRFC is contrasted to traditional machine learning methods and deep feature extractors. The results of the experiment indicate that DRFC achieves 97.8% accuracy and 97.4% F1-score on Fruits-360, and 92.6% accuracy with a macro-F1 of 91.3 on USDA CDL/ BigEarthNet, and costs less in terms of runtime and memory usage than the baseline algorithms by a factor of 2–3. These results show that DRFC is a useful, scalable, and computationally efficient solution for cloud-based agricultural image analytics, mainly when big deep learning models cannot be effectively used due to resource limitations.
Fu et al. (Wed,) studied this question.