Abstract: This paper describes a sophisticated deep learning system aimed at examining seed images and predicting several agricultural parameters from a single image input. The system uses a multi-task learning ResNet18-based architecture that can classify seed type, evaluate viability, estimate growth rate, examine surface texture, and predict environmental conditions like temperature, humidity, moisture content, and light intensity simultaneously. In contrast to conventional systems that depend on physical sensors or lab equipment, this method uses only visual information, and hence it is a cost-efficient and scalable solution for smart agriculture. The model shows high accuracy in all tasks and is incorporated into a real-time user interface, allowing for instant and useful application in the field. This paper brings out the capabilities of computer vision and deep learning in revolutionizing traditional farming methods into smart, sensorless systems.

| DOI: 10.17148/IARJSET.2025.125219