Portable high-throughput seed sorting using smartphone cameras

Overview

This innovative solution offers a portable and low-cost seed sorting system that utilizes the advanced capabilities of commodity smartphones. By employing both front and back-facing cameras, along with additional imaging technologies such as NIR and LiDAR, the system classifies seeds accurately and rapidly. This approach addresses the limitations of traditional seed sorting devices, which are often large, expensive, and lack adaptability.

Technical specifications

The system features a modular, 3D-printed tabletop structure designed to funnel seeds towards the smartphone cameras. This setup takes advantage of the multiple high-resolution cameras and powerful processors in smartphones, enabling real-time seed classification using deep learning models like DeepSort and SeedSortNet. Key technical components include:

• Multi-camera integration: Simultaneous use of front and back cameras for multi-angle viewing.
• Advanced imaging: Capability to capture RGB, NIR, hyperspectral, and LiDAR data streams.
• Deep learning models: Custom and pretrained models for rapid classification and segmentation.
• Adaptability: Easily updated software to handle new seed types and traits.
• Real-time processing: Utilizes smartphone AI accelerators and GPUs for on-device computation.

Technology readiness level

Currently at TRL 2, the technology is in the early stages of development. Future validation plans include optimizing the multi-modal deep learning model, integrating the system in various environmental conditions, and enhancing the modular housing for increased portability and efficiency. These efforts aim to address current challenges and push the technology towards higher readiness levels for practical deployment.