Weed Detection in Spinach Fields Using Hyperspectral Snapshot Imaging

Weeds are a persistent and costly challenge for farmers. Competing with crops for vital
resources like nutrients, water, and sunlight, they can dramatically reduce yield and quality
leading to substantial economic losses. As the demand grows to maximise productivity whilst
minimising costs and environmental impact, new solutions are needed to support targeted
weed control strategies.

Why traditional methods fall short

Weeds often resemble crops in shape, colour, and size, especially in early or dense growth
stages—making them hard to identify using traditional methods. Historically, farmers have
relied on manual labour and chemical herbicides to manage and control weeds. However,
they came with limitations:

• Manual removal was time and labour-intensive.
• Chemical herbicides pose environmental and health risks.
• Chemical overuse promotes herbicide resistance.
• Uniform spraying wastes resources without precise weed location data.

A sustainable alternative: Hyperspectral Imaging

As the demand for sustainable weed management increased farmers have begun to seek
ways to reduce herbicide use whilst maintaining crop health and productivity.

To meet these challenges, hyperspectral imaging (HSI) offers a promising path forward.
HSI captures the unique spectral response of plants across hundreds of wavelengths.
Unlike RGB cameras, HSI can detect subtle chemical differences in vegetation.

What HSI enables:

• Accurate weed identification;
• even in complex environments.
• Data-driven, site-specific weed control.
• Reduced herbicides use & operational costs.
• Real-time or near real-time field monitoring.
• High model performance with minimal
• training data.

Identifying Weeds in Spinach Fields

To demonstrate the utility of HSI in the field, a proof-of-concept (POC) was conducted to
detect a specific weed species growing alongside mature spinach. This task posed a tough
challenge due to their visual similarity. Using just a few seconds of video, from a few selected
frames data was extracted, L2 normalised and used to train a spectral classifier. The model
achieved 74% accuracy on previously unseen data, a strong result given the minimal input
and the visual similarity between crop and weed.

Field Set Up
Camera	Configuration:	Capture rate:
Living Optics hyperspectral snapshot camera	Handheld, walking at approximately 1 m/s	30 frames per second

Field-Ready Precision Weed Management

The POC highlighted the value of the Living Optics hyperspectral imaging camera in
real-world agricultural settings. With minimal training data, real-time capability, and solid
performance, this technology is poised to make site-specific weed management smarter,
faster, and more sustainable.

To learn more contact us today.