Deep Learning Based Visual Seed Quality Inspection and Plant Disease Prediction

Abstract

This article presents an Agri-informatics framework that applies artificial intelligence to smart and sustainable farming, focusing on Zea mays seed health and yield forecasting. This work intakes a dataset of$9,000 \text{RGB}$seed images to gather simple geometric measures like aspect ratio, area, and circularity. A lightweight neural network has been trained using these parameters, capable of performing both classification and regression tasks. The classifier identifies whether seeds are healthy or diseased, while the regressor predicts yield potential on a scale from 1 to 100. Validation experiments demonstrated about 62% accuracy for classification and a mean squared error of roughly 37.9 for yield estimation. To make results easier to understand, Grad-CAM-inspired simulations were performed to show how each feature affects the decision-making process. This method shows that lightweight, non-destructive models can provide reliable insights into seed health without losing computational efficiency. The system can run on edge devices, IoT platforms, or be integrated into digital agriculture pipelines. Ultimately, it offers a scalable and straightforward AI-based decision framework for precision farming and modern seed monitoring systems.

Plant phenotyping relevance

種子画像から形態特徴を抽出し、健康状態の分類と収量ポテンシャル推定を行うAI手法が研究の中心であり、植物表現型の取得・推定方法に該当する。

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.