Disease detection & pod counting with AI.
Our AI crop disease detection combines computer vision, machine learning, large language models, satellite data, weather data and geospatial information — integrated into HSAT's global crop survey capability of ground teams, specialised apps and scalable analysis.
AI reports on every pod we photograph.
Each field photograph is analysed by our AI, which returns a plain-language report identifying the suspected disease, the evidence it saw, and a risk score out of ten. Healthy pods are flagged with low scores; disease signatures like Black Pod (Phytophthora spp.) score high.
Works on low-quality field images.
The disease analyses below are from low-quality images captured in Ivory Coast by our local ground team. Alternatively, client staff can use our app to capture the same content — the AI grades whatever imagery comes in.
The same images count pods, too.
The same images we use for disease detection are also used to count pods with AI. This output was achieved with computer vision and only 10 minutes of training — accuracy improves considerably over time as the model sees more field data.
Human counts, at any scale.
The same images can also be counted manually. For small-scale work HSAT does this internally using our offshore team. For large-scale projects we deploy a trained crowd of 5,000+ through the Mechanical Turk (MTurk) process, completing the counting in days rather than weeks.
AI vs human: the harder end of the dataset.
The images below were gathered in Ghana and used to develop our pod-counting AI model. They were intentionally selected as the pods are less obvious than the norm. A large number of such images were counted by both humans and AI — below are the two passes over the same three frames.
On Image C, AI identified a pod the human reviewer missed (arrow overlay). This is characteristic of the AI's value at scale: it doesn't tire, and it reviews every frame with the same attention.
A full stack, not a single model.
Disease detection and pod counting aren't standalone demos — they plug into HSAT's wider crop intelligence pipeline.
Ground teams
Local surveyors and client staff capture field photos via our specialised apps — every pod tagged to its exact GPS location.
AI analysis
Computer vision, machine learning and large language models review every image — returning disease risk scores and pod counts with plain-language evidence.
Human QA
Offshore teams handle small jobs; a 5,000+ MTurk workforce handles large ones. Counts cross-checked against the AI pass.
Fused output
Field data is fused with satellite imagery, weather and geospatial layers to produce forecasts, disease maps and risk intelligence.
Put disease detection to work on your crop.
Whether you need a one-off diagnostic run or a continuous monitoring programme across a region, we can scope it.