Phinorm
Phinorm provides a software upgrade for existing sensing hardware—radar, acoustics, visual, and vibration systems—that improves detection accuracy through a trade-off-free mathematical approach. The solution delivers 2–4× fewer classification errors in small-sample radar classification and adapts rapidly to changing environments while remaining SWaP-efficient for embedded and edge deployments. It integrates with existing machine-learning stacks and can be reused across multiple sensor modalities.
- Defense Tech
- Artificial Intelligence
- Internet of Things
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Neural networks often struggle to adapt quickly when operating conditions shift—such as changes in environment, platform, clutter, or sensor swap—leading to degraded detection performance. In small-sample settings, retraining costs time and compute, which is especially problematic in time-critical defense scenarios where every second of warning matters.
Solution
Phinorm provides a software upgrade that enhances the detection capabilities of existing sensing hardware such as radar, acoustics, visual, and vibration systems. By applying a trade-off-free mathematical approach originating from a long-forgotten branch of mathematics, the solution delivers 2–4× fewer classification errors versus scientific benchmarks in small-sample radar classification. It adapts rapidly to changing environments while remaining compute-efficient and deployable on any device, including embedded and edge platforms. The approach complements rather than replaces existing machine-learning stacks, integrating as a drop-in upgrade to sensing pipelines. It is multi-modal, allowing the same approach to be reused across radar, acoustics, visual, and vibration sensor types.
Target Audience
Primary customers are defense and security organizations, including air defense systems, as well as hardware partners seeking to turn commodity sensors into premium products. The solution is also relevant for NATO-aligned research labs and data partners working on UAV detection and multi-sensor fusion.
Features
- Trade-off-free mathematical approach that improves accuracy without sacrificing SWaP efficiency, suitable for embedded and edge deployments
- 2–4× fewer classification errors versus scientific benchmarks in small-sample radar classification (e.g., 94% accuracy vs. 76% for best reference model in bird vs. drone classification)
- Rapid adaptation to changing environments, converging quickly to new settings without costly retraining
- Integrates with existing ML stacks as a drop-in software upgrade for sensing pipelines
- Multi-modal support across radar, acoustics, visual, and vibration systems, enabling reuse of the approach across sensor types
- Demonstrated results: 45% fewer errors on detecting UAVs from audio data and 75.5% more accurate detection on radar data