Bi-GRU Attention-CapsNet with Parallel Spatio-Temporal Attention-TCN Based Smart Agricultural System for Crop and Fertilizer Recommendation with Weed Detection
Optoelectronics, Instrumentation and Data Processing · 26.10.2026
DOI: 10.3103/s8756699026700561 ↗Kurz gefasst
Fachartikel in der Zeitschrift Optoelectronics, Instrumentation and Data Processing, online erschienen am 26.10.2026. Verfasst von Sukham Romen Singh und 1 weiteren Autor:innen. Die Arbeit nennt 27 Literaturangaben.
Abstract (Originalsprache)
Intelligent monitoring systems using Internet of Things (IoT) are progressively penetrating the agricultural sector to make automated decisions for enhancing crop growth, fertilizer utilization, and weed control. However, most of the existing AI models are not effective in dealing with missing values, low-resolution images, and dynamic changes of features over time. To address these shortcomings, a smart agricultural system for crop and fertilizer recommendation with weed detection is developed using attention-enabled deep learning models. For the recommendation module, soil condition, weather, and other related characteristics were gathered and processed. Pre-processing of the data is achieved using InfoGAN, Ordered Quantile normalization (ORQ), and Isolation Forest for missing value imputation, range normalization, and outlier detection. Then, the Deep Factorization Machine (DFM) is used to extract essential features. The extracted features are processed with the Bi-GRU Attention-CapsNet (Bi-GAC) module, where appropriate crops and fertilizers are recommended through captured sequential dependencies. In the case of weed monitoring, the images captured through IoT-based surveillance are improved with the help of Super-Resolution-Cycle Generative Adversarial Network for high resolution, Minimal Color Loss and Locally Adaptive Contrast Enhancement (MLLE), and Improved NLM filtering for noise removal. The EfficientNetV2 is used to extract weed features, and Parallel Spatio-Temporal Attention-TCN (PSTA-TCN) is used to detect weed based on spatial-temporal patterns. The proposed framework attains accuracy values of 99.36 $$\%$$ for crop and 97.80 $$\%$$ for fertilizer. Precision is 97.30 $$\%$$ for weed detection. According to the results, the proposed model has effective recommendations and weed detection for supporting a precise and strong solution to precision agriculture.
Zitieren
APA
Singh, S. R., & Ramasamy, V. (2026). Bi-GRU Attention-CapsNet with Parallel Spatio-Temporal Attention-TCN Based Smart Agricultural System for Crop and Fertilizer Recommendation with Weed Detection. Optoelectronics, Instrumentation and Data Processing. https://doi.org/10.3103/s8756699026700561
Vancouver
Singh SR, Ramasamy V. Bi-GRU Attention-CapsNet with Parallel Spatio-Temporal Attention-TCN Based Smart Agricultural System for Crop and Fertilizer Recommendation with Weed Detection. Optoelectronics, Instrumentation and Data Processing. 2026. doi:10.3103/s8756699026700561
BibTeX
@article{Singh2026,
author = {Singh, Sukham Romen and Ramasamy, Viswanathan},
title = {Bi-GRU Attention-CapsNet with Parallel Spatio-Temporal Attention-TCN Based Smart Agricultural System for Crop and Fertilizer Recommendation with Weed Detection},
journal = {Optoelectronics, Instrumentation and Data Processing},
year = {2026},
doi = {10.3103/s8756699026700561}
}
EUVIMED