Crop Yield Prediction using Hyperspectral Imagery and Machine Learning Algorithms Deployed on Edge Computing Nodes

Abstract

This research work proposed the precision agriculture is supported by accurate future crop yield forecasting which allows optimal use of resources automated use of crops and planning production of food. The problems that conventional predictive systems basing on cloud computing are likely to face include longer latency unreliable network connectivity and heavy processing power requirements which are especially troublesome in the rural farm environment. To deal with these drawbacks the study proposes an Edge-AI-driven framework of crop yield estimation which entails the integration of hyperspectral imaging with real-time measurements of agricultural sensors of soil moisture, pH level, temperature, and humidity. it is based on an STM32 Edge-AI microcontroller and a hybrid deep learning architecture Conv LSTM-ViT (Convolutional Long Short-Term Memory embedded with Vision Transformer) to process spectral changes and environmental changes over time and identify the factors that influence crop growth. The model is able to make decisions in a short time track continuously and lessen cloud reliance by executing inference at the edge. The proposed Conv LSTM-ViT Edge AI model outperforms Random Forest, XG Boost, CNN, and Conv LSTM with up to 14% higher accuracy and 70–80% lower inference latency, demonstrating strong suitability for real-time smart agriculture deployments. Metadata and prediction outcomes are safely uploaded to the Blynk IoT cloud to be visualized and offer decision support to the farm levels. Field testing applications with UAVs further support that the system is energy-saving scalable and able to assist real-time smart farming processes which creates a sustainable system of future farming technologies.

Plant phenotyping relevance

ハイパースペクトル画像とエッジAIによって作物収量を推定する取得・解析手法を開発し、既存モデルとの性能比較と実地検証を行っており、収量推定法が中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.