Crop diseases are a major cause of yield loss and are hardest to manage when caught late. For many smallholders, the nearest plant pathologist or KVK scientist may be some distance away, and by the time an expert sees the crop, a disease may have spread.
Machine learning, and image-based AI in particular, offers a way to get a first opinion quickly from a smartphone photo.
How image-based diagnosis works
Modern image recognition models learn by example. During training, a model is shown large numbers of labelled images, for instance leaves with early blight, leaves with bacterial spot, and healthy leaves. Over time it learns the visual patterns associated with each condition: the shape and colour of lesions, the presence of concentric rings or halos, powdery growth, mosaic patterns, wilting, and so on.
When you upload a new photo, the model compares what it sees with the patterns it has learned and produces its best assessment. Newer multimodal AI models can also combine the image with text context, such as the crop name, and explain their reasoning in plain language.
Two points are important to understand:
- The output is a likelihood, not a certainty: the model gives its most probable explanation based on the image.
- Performance depends on the image: a model can only judge what is visible in the photo. Images taken in real fields, with mixed backgrounds and uneven light, are harder than the clean images often used in research datasets.
What makes a good photo
Photo quality has a large effect on the result. When photographing a plant for diagnosis:
- Focus on the symptom: get close enough that lesions, spots or discolouration are clearly visible and sharp.
- Use natural, even light: avoid harsh midday glare and deep shadow. Light shade or an overcast sky works well.
- Show both sides if relevant: some diseases and pests are more visible on the underside of the leaf.
- Include context in a second photo: a wider shot of the whole plant or row can show wilting patterns, distribution and plant stage.
- Pick a representative sample: photograph a leaf that shows typical symptoms, not just the single worst one.
- Keep the background simple and avoid blurry or dark images. If the photo is unclear to you, it will be unclear to the model.
Limits and look-alikes
Many crop problems look similar, and some are not diseases at all. Common sources of confusion include:
- Nutrient deficiencies, such as nitrogen, potassium or zinc deficiency, which can cause yellowing or leaf margin burn similar to some diseases
- Herbicide drift or chemical injury, which can produce spots, curling or scorching
- Insect and mite damage, where feeding marks or virus symptoms spread by insects can resemble fungal or bacterial disease
- Environmental stress, including heat, waterlogging, cold injury or sunscald
- Multiple problems at once, where a plant weakened by one issue is then attacked by another
Very early infections may also show few visible symptoms, which limits what any photo-based tool can detect.
Why confirmation with a KVK or agronomist matters
An AI diagnosis is a fast first opinion. Before spending money on treatment, especially chemical treatment, it is good practice to confirm with your local Krishi Vigyan Kendra, an extension worker or a qualified agronomist. They can examine the plant in hand, check the field, consider local disease history and recommend products that are approved and available in your area. They can also send samples for laboratory testing when needed.
The wrong treatment wastes money, may harm beneficial insects, and leaves the real problem uncontrolled.
Weather-driven disease risk
Images show diseases that have already appeared. Weather data can help anticipate them. Many fungal and bacterial diseases depend on moisture: high humidity, prolonged leaf wetness and moderate temperatures over several days create favourable conditions for infection and spread.
When such conditions are forecast, a practical response is to scout fields more often, focus on susceptible crops and varieties, and, where recommended by local experts, consider preventive measures. Combining weather-based risk with photo-based diagnosis helps catch problems earlier.
Research directions
Research on earlier and more precise disease detection is active. Some directions include:
- Hyperspectral imaging, which captures many narrow wavelength bands and may detect changes in plant chemistry before symptoms are visible to the eye
- Thermal imaging, which can reveal changes in leaf temperature linked to infection or water stress
- Drone and satellite monitoring, which can map unusual patches across large areas for targeted field checks
- Combining data sources, such as images, weather and field history, to improve accuracy
Most of these approaches are still being refined and are not yet practical or affordable for routine smallholder use.
How Sodagro's Crop Disease Scan works
Sodagro's Crop Disease Scan lets you upload a photo of an affected plant. The AI returns:
- A likely diagnosis
- An assessment of severity
- Treatment options, including organic and chemical approaches
- A recommendation to confirm the diagnosis with your local KVK or an agronomist before acting
For follow-up questions, the AI Agronomy Advisor can use your field details and the local 7-day forecast to help you think through next steps. Sodagro's weather alerts also flag fungal-disease risk when several humid days are forecast for a field. You can see how these tools fit together on our Platform page.
Try Sodagro for your fields
Ask questions about your crop, check a sick plant from a photo and see weather alerts for your fields. You can start for free.