Distributed image analysis platform for greenhouse fertigation decision support
A software system that uses edge computing devices to extract features from time-series images of tomato plants, enabling more responsive and translatable fertigation decision support than traditional models. Designed for distributed deployment across greenhouse operations with integration of meteorological and input data.
Overview
This solution addresses a critical challenge in precision agriculture: the difficulty of obtaining field-specific data in near real-time to support fertigation decisions for tomato crops. While existing decision support models such as GesCoN provide robust recommendations, their practical utility is limited by data collection challenges and reduced translatability to novel growing conditions due to overfitting.
The platform enables distributed image collection across wide greenhouse areas, with features extracted on edge computing devices and communicated to a central fertigation decision support system via wide area networks. By coupling image-derived features with traditional input and meteorological data, the system offers more responsive and adaptable decision support tools suitable for diverse growing environments.
Technical specifications
Core capabilities:
- Edge-based feature extraction — Software deployed on edge computing devices processes time-series images of tomato foliage locally, reducing bandwidth requirements and enabling distributed deployment
- Time-series image analysis — Adapted from validated minirhizotron image analysis algorithms that specialize in discovering differential features across sequential images
- Stress detection sensitivity — Preliminary Phase II SBIR results demonstrate that time-series derived image features are sensitive to environmental stressors and can differentiate experimental groups within pathological inoculation experiments
- Data integration — Combines image-derived features with known agricultural inputs and meteorological data to develop comprehensive decision support models
- Microservice architecture — Built using modern, scalable microservice designs with test-driven development practices for reliability and maintainability
- Flexible deployment — Supports cloud infrastructures as well as secure, isolated systems
Technology readiness level
The underlying image analysis methodology has been validated through preliminary results from an ongoing DOE-funded Phase II SBIR involving minirhizotron image analysis. These results confirm that time-series derived image features are sensitive to environmental stressors and capable of differentiating experimental groups.
The proposed adaptation to above-ground foliage imaging for fertigation decision support represents a logical extension of this validated technology. Formal validation will involve model sensitivity analysis of image-derived features combined with qualitative assessment of feasibility for in situ image collection and transmission. The development team is actively seeking field trial partners to collect image and yield data for model training and validation, representing a clear pathway toward commercial deployment in greenhouse operations.
About Geometric Data Analytics
Geometric Data Analytics (GDA) is a Durham, North Carolina-based research, development, and consulting company that specializes in solving complex data analysis problems. The firm is built upon expertise in topological data analysis, applied mathematics, machine learning, and software engineering. Their team develops custom algorithms and software architectures, often working in domains where standard off-the-shelf artificial intelligence and machine learning solutions are insufficient. By utilizing test-driven development and modern, scalable microservice architectures, GDA provides interoperable and maintainable technical solutions designed for deployment across diverse environments, including cloud infrastructures and secure, isolated systems.
The company serves clients in the government, military, and commercial sectors, offering capabilities in areas such as anomaly detection, high-dimensional data analysis, agent-based modeling, and signal processing. GDA focuses on delivering scientific research and algorithmic development that can be seamlessly integrated into larger systems. Their methodology emphasizes speed and reliability, enabling partners to progress from theoretical concepts to functional, deployable prototypes efficiently. GDA also supports open-source initiatives and provides consulting to help organizations modernize their development pipelines through CI/CD practices and containerized deployment technologies.
delivered to your inbox.