Comment

Crop health ยท App test guide

Plant Disease App: How to Test One on Your Own Crop

Choose a plant disease app by testing it on your own crop. One model scored 99.35% on controlled photos and about 31% on other images (Mohanty et al., 2016). Agrio, PlantVillage+ and PictureThis suit different growers.

Authored by Farmonaut Research & Content Team
Reviewed and verified by Farmonaut Agricultural Experts Team
Updated 5 November 2026 ยท 15 min read
Sentinel-2 Satellite View Of Vegetable Fields In The Salinas Valley, California
Sentinel-2 view of Vegetable fields in the Salinas Valley, California. Contains modified Copernicus Sentinel data, processed in Google Earth Engine.

Key takeaways

  • One published model’s accuracy fell from 99.35% on controlled photos to about 31% on other images (Mohanty et al., 2016).
  • In a 2020 cassava trial, using six leaves instead of one upper leaf raised mosaic disease diagnosis from 53% to 93%.
  • Mild symptoms were hardest: 13% to 37% accuracy in the same trial.
  • A 10-case blind test on known problems shows whether a tool knows your crop.
  • University lab fees listed in October 2026 ran from US$20 to US$35 per in-state sample at the three labs shown.
In this article
  1. Which plant disease app fits which grower
  2. How a plant disease identifier app reaches its answer
  3. How accurate is a plant disease app in real use?
  4. Test any plant disease app in five steps
  5. What a field trial shows about taking better photos
  6. Where a photo diagnosis is most likely to fail
  7. What satellite maps add to a phone diagnosis
  8. Act on the app's answer or confirm first?
  9. When to use a lab instead of an app
  10. Why an early, correct answer is worth the effort
  11. Six checks before you pay for a plant disease app
  12. How we checked these tools
  13. FAQ

The best plant disease app is the one that gets your own crop right when you test it, because accuracy can fall sharply on photos unlike the ones a model was trained on. One well-known model scored 99.35% on controlled photos and about 31% on images from other sources (Mohanty et al., 2016). For a quick name on a sick plant, try Agrio, PlantVillage+ or PictureThis, then run the short test below before you spray on the answer.

This guide is about trust, not just features. It gives a five-step test you can run on known problems, often within about an hour, photo rules backed by a field trial, and a calculator that shows when a lab check costs less than acting on a wrong answer. If you want a longer side-by-side of apps and prices, see our comparison of farm disease apps.

Which plant disease app fits which grower

Store pages show three kinds of tool. Some name a disease from a photo. Some name the plant species, which is a different job. Some are expert reference sites with no photo feature at all. Prices below are US App Store list prices as shown on 15 October 2026. Check the store for your country before you pay.

Tool Best for What the developer lists Listed cost
Agrio Growers and crop advisors Photo diagnosis, IPM protocols, field weather, NDVI maps, scouting reports Plans from US$4.49 to US$349.99 in the store list; period varies by plan
PlantVillage+ Field staff who also want records Photo identification of possible problems, advice, a field journal No charge listed; Penn State shown as developer
PictureThis Home gardens and houseplants Plant naming, sick-plant diagnosis, care reminders Pro from US$3.99 to US$39.99 in the store list
Pl@ntNet Naming an unknown plant Species identification from a picture Not a disease tool
Crop Protection Network US and Canadian field crops Web tools and publications from land-grant universities Reference, not a photo app

Agrio

Agrio’s App Store page aims at crop advisors, farmers and gardeners. It adds satellite indices, weather warnings and shareable scouting reports to the photo tool, so it suits someone who walks fields for a living. The store list includes “expert response tokens” at US$19.99, a useful route when the photo answer looks doubtful. Its Google Play page showed 500K+ downloads and a 4.2 rating in October 2026, and it notes ads and in-app purchases.

Best for: advisors and farms that want diagnosis and scouting records in one place.

PlantVillage+

This app grew out of Nuru, and the store listing still uses the Nuru package name. Its Google Play page says AI results should support, not replace, local expertise and qualified advice. That is a sound rule for any tool on this list. It also says features vary by crop, country, language and connection, so check that your crop is covered first. The original Nuru model was built around cassava, which limits how much of its trial data you can apply to other crops.

Best for: crew leaders and extension staff who need field records as well as a first opinion.

PictureThis

PictureThis is mainly a plant naming and care app with a very large user base: 1.1 million ratings at 4.8 on the US store. Its “over 98% accuracy” line is the developer’s own claim and sits with plant identification. It is not a field-tested figure for crop disease. Use it for roses, tomatoes in a raised bed and houseplants rather than for a decision on a large block.

Best for: home growers and small market gardens.

Pl@ntNet and reference sites

People often install Pl@ntNet when they want a disease answer. It is a citizen science project that names plants from pictures, so it helps with an unknown weed or volunteer but will not diagnose your crop. For US and Canadian field crops, the Crop Protection Network is a product of land-grant universities and publishes fungicide efficacy tables and disease loss estimates. It works well as the second opinion you read after a photo tool gives you a name.

How a plant disease identifier app reaches its answer

Photo tools generally follow the same path. You take a picture. A trained image model compares the visible pattern, such as spots, yellowing or curling, with the examples it learned from. It returns one or more likely names, often with treatment notes. Some tools then add weather or location data to judge the risk.

Three limits follow from this design, and no feature list will tell you about them:

  • It can only name what it was taught. A disease or variety missing from the training photos has no correct answer available, and the tool will still offer its nearest match.
  • It sees symptoms, not causes. Herbicide drift, nutrient shortage and root problems can look like leaf disease in a photo.
  • Training photos may not look like your field. A review in Frontiers in Plant Science (2019) reported that 62.7% of images in a later 87,848-image version of the widely used PlantVillage dataset were taken under controlled conditions.

How accurate is a plant disease app in real use?

Treat any single accuracy score in an app comparison with care, and ask what photos it was tested on. The useful evidence comes from peer-reviewed tests of the underlying models, and those tests explain why you should run your own check. Mohanty and colleagues trained a model on 54,306 leaf images covering 14 crops and 26 diseases. It scored 99.35% on held-back images from the same set. On two small sets of images from other sources, it fell to about 31%, while the correct answer was in its top five guesses 53% to 66% of the time.

Chart

One study, different photos

Share of images correctly labelled, best model setup for each set, top answer; other sets were 121 and 119 images

0.0%25.0%50.0%75.0%100.0%99.3%Controlled photos31.4%Other images, set131.7%Other images, set22.6%Random guessSource: Mohanty, Hughes and Salathรฉ, Frontiers in Plant Science, 2016.
0.0%25.0%50.0%75.0%100.0%Controlled photos99.3%Other images, set 131.4%Other images, set 231.7%Random guess2.6%Source: Mohanty, Hughes and Salathรฉ, Frontiers in PlantScience, 2016.
The same kind of model that was near-perfect on controlled photos got about three in ten right elsewhere.

That study is from 2016 and models have moved on, so treat the gap as the lesson rather than the exact number. The authors point to the cause: their photos showed single leaves, facing up, on a plain background. A real plant in a real field often looks quite different. They also found that telling the model the crop in advance raised accuracy on the outside images to between 41% and 55%, though random guessing also improves on those smaller crop-limited sets (18% to 31%). It is still sensible to select your crop if the app offers that option.

Look at the top-five figure too. A list of likely names is far more useful than a single confident answer, so read past the first result and compare each name against a trusted fact sheet.

Test any plant disease app in five steps

This test costs little and should take about an hour, depending on how many cases you have ready. It tells you more about your crop than any store rating, since a rating measures how much people like an app, not whether it is right.

  1. Collect known cases. Gather about 10 plants or photos where the cause is already settled: a past lab report, an advisor’s visit, or a clear case your agronomist will confirm. Include two or three healthy plants and one problem that is not a disease, such as herbicide injury.
  2. Run them blind. Have someone else record the true answers. Photograph each case in normal field light and enter the results in a table.
  3. Score three things. Was the first answer right? Was the right answer anywhere in the list? Did the app wrongly call a healthy plant sick?
  4. Check the treatment notes. Is the product registered for your crop where you farm? Advice written for another country may not be legal for you.
  5. Repeat at a different stage. Mild symptoms were the hard ones in the cassava trial, so run a second small batch on mild cases when you can.

Ten cases is a small sample. It will not give you a true accuracy rate. It will show whether the tool knows your crop at all, and it will expose an app that labels everything as disease. Decide your own pass mark in advance and write it down, so a slick interface does not change your mind afterwards.

What a field trial shows about taking better photos

The best public field evidence is the 2020 trial of PlantVillage Nuru on cassava in East Africa, published in Frontiers in Plant Science. It compared the app with cassava experts, extension officers and farmers, and checked experts against lab tests. It covers one crop and three problems, so do not assume the same scores for your crop. The practical lessons on photos carry over well.

The biggest one is this: more leaves meant far better diagnosis. With one upper leaf, the app scored 53% on cassava mosaic disease. With three upper and three lower leaves, it reached 93%, the same score the paper gives for two leaves on mosaic. The paper’s abstract gives 74% to 88% for six leaves across conditions.

From this trial and the Mohanty limits above, a simple photo routine follows:

  • Photograph several leaves from each plant, both high and low in the canopy.
  • Sample several plants, not the single worst one.
  • Include the edge of a symptom, where sick tissue meets healthy tissue.
  • Shoot stems, fruit and the underside of leaves if the symptoms are there.
  • Keep the leaf in focus, in even light, without a harsh shadow across it.
  • Note the crop, variety, growth stage, recent weather and last spray. The app may not ask, but you will need them later.

Chart

More leaves, better diagnosis

PlantVillage Nuru diagnosis rate on cassava plants, by number of leaves used

One upper leafThree upper + three lower0%25%50%75%100%Mosaic disease53%93%Green mite damage67%93%Brown streak disease0%73%Source: Mrisho et al., Frontiers in Plant Science, 2020.
One upper leafThree upper + three lower0%25%50%75%100%Mosaic disease53%93%Green mite damage67%93%Brown streak disease0%73%Source: Mrisho et al., Frontiers in Plant Science, 2020.
Each condition improved with six leaves; brown streak rose from zero on one upper leaf.

Where a photo diagnosis is most likely to fail

Symptoms were not equally easy. In the same trial, in-field accuracy on single leaves ran from 90% for healthy leaves down to 21% for cassava brown streak disease, whose foliar signs are often patchy. Mild cases were weak: the app identified mild symptoms with 13% to 37% accuracy, against 27% to 83% for intermediate symptoms. This matters if mild symptoms are what you see first, when you may most want a reliable answer.

Chart

Single-leaf accuracy in the field, by condition

PlantVillage Nuru, in-field identification on cassava leaves

0%25%50%75%100%No symptoms90%Mosaic disease59%Green mite damage56%Brown streak disease21%Source: Mrisho et al., Frontiers in Plant Science, 2020.
0%25%50%75%100%No symptoms90%Mosaic disease59%Green mite damage56%Brown streak disease21%Source: Mrisho et al., Frontiers in Plant Science, 2020.
Healthy leaves were recognised far more reliably than any of the three problems.

People need context too. In the on-screen symptom test, trained researchers averaged 86%, trained extension officers 49% and trained farmers 23%. Across the study’s assessment, Nuru scored 65% in 2020, above extension agents at 40% to 58% and farmers at 18% to 31%, but below the trained researchers. So a good tool can beat a non-specialist on its own trained crop, and a real specialist remains the better check. These are cassava results and should not be read as a general ranking of apps against people.

What satellite maps add to a phone diagnosis

A phone tool answers “what is on this leaf?” It says nothing about how far the problem has spread, or where you should walk next. Satellite vegetation maps work from the opposite end. They show parts of a field that are changing compared with the rest, but they cannot name a disease, since drought, nutrient shortage and pests can all give a similar signal.

The two fit together on larger farms. Use the map to pick where to scout, the phone at that spot for a first opinion, and a lab or advisor for anything costly. Farmonaut provides satellite vegetation maps for this first step; it does not diagnose a named disease from space. Agrio also lists NDVI maps within its app, so compare what each offers against the size of your fields.

A walkthrough of satellite crop monitoring, including how to read NDVI for your own field.

โ–ถ A walkthrough of satellite crop monitoring, including how to read NDVI for your own field. ยท 19:36

Act on the app’s answer or confirm first?

A wrong answer has a real price: a spray that was never needed, plus the delay in treating the true cause. A lab check also has a price, in fees, shipping and days of waiting. The calculator compares the first with the second for one decision. Use your own figures; the defaults are examples set close to break-even. It leaves out yield lost to a wrong or late treatment, because that depends on the disease, so also weigh that risk for fast-moving problems.

Try the numbers for your own farm below.

Calculator

Is a lab check cheaper than a possible wrong spray?

Enter your own field, spray cost, doubt level and lab costs.

acres

US$/acre

%

samples

US$/sample

US$/sample
$0expected spray waste avoided minus lab cost (US$); above zero favours checking
  • Expected wasted spray if you act now (US$): $48
  • Cost of the lab check (US$): $48
  • Chance of error where the two costs match: 40.0 %

Example values only, near break-even, not a benchmark or forecast. All prices are inputs: use today’s price. The chance of a wrong answer is your own judgement from your test results; no general figure exists. Sample fee example reflects the US$35 routine fee listed by Virginia Tech and Texas labs in October 2026. Yield loss from a wrong or late treatment is not included. Follow the product label and local rules.

When to use a lab instead of an app

Send a sample or call an advisor when the decision is expensive, the symptoms are unusual, or the first treatment did not work. Fees at the three university labs below were under US$100 per routine sample, as listed on their pages in October 2026:

Lab Listed routine fee Worth knowing
Virginia Tech US$35 in-state, US$70 from other US states Accepts digital, physical or combined samples
Texas A&M US$35 per sample US$20 per-sample surcharge from out of state
University of Rhode Island US$20 per sample Molecular identification may cost extra

See the Virginia Tech clinic’s sample guide, the Texas lab’s fee schedule and the Rhode Island lab’s page for current terms. In the US, the National Plant Diagnostic Network is a group of labs across all states and territories, with a map to find yours. Its site, viewed in October 2026, said funding delays meant network activities would be paused or run at reduced capacity from 1 September 2026, so call your lab before you ship. UK home gardeners who are members can use RHS Gardening Advice.

Some problems carry legal duties, whatever an app says. In England, UK government guidance says that if you find or suspect a quarantine organism you must contact your local Plant Health and Seeds Inspectorate office immediately. Canada, Australia, New Zealand and the EU have their own reporting rules, so check with your national plant health agency.

Before any treatment, check the product label and local rules, and get professional advice if you are unsure. An app’s suggestion does not make a product legal or safe for your crop.

Why an early, correct answer is worth the effort

Disease pressure is expensive at scale. An expert survey in Nature Ecology and Evolution (Savary et al., 2019) estimated global yield losses to pathogens and pests of 21.5% for wheat, 30.0% for rice, 22.5% for maize, 17.2% for potato and 21.4% for soybean. These figures combine diseases and pests and are expert estimates, not field measurements, but they show the stakes. The Crop Protection Network’s soybean disease loss estimates give a closer view for the US and Ontario.

A correct answer has benefits that go beyond the current field. It helps you avoid a wasted spray, choose the right product group and rotate resistant varieties next season. It also builds a record you can compare year after year. For the wider plan around a diagnosis, see our guide to plant disease management strategies.

A guide to common plant diseases and pests and how growers can spot problems early.

โ–ถ A guide to common plant diseases and pests and how growers can spot problems early. ยท 17:29

Six checks before you pay for a plant disease app

  • Crop coverage. Is your crop named, or just “vegetables”? Ask the developer for a list if it is not shown.
  • Offline use. If your fields have poor signal, test it in the field with data turned off before you rely on it.
  • Answer format. Prefer a tool that shows several possible causes and how sure it is over one that gives a single flat answer.
  • Local treatment advice. Advice should match the products registered where you farm.
  • Access to a person. Can you send a doubtful case to an agronomist, and what will that cost?
  • Data terms. Photos can carry location data. Read who may use your images and field details, and whether you can delete them.

Also watch auto-renewal. PictureThis, for example, states that its subscription renews unless turned off at least 24 hours before the period ends. Run your test during any trial period, and note the end date.

How we checked these tools

Each tool here was checked against its own store page or website in October 2026. The accuracy discussion relies on peer-reviewed studies, each named with its crop and date. Those studies tested models and one named app, so they do not rank the tools against each other. Store ratings appear for context, not as a measure of accuracy. Features and prices change often, so confirm them in the store before you buy.

Related pages on our site: satellite and weather API documentation, carbon footprint monitoring, product traceability and fleet management; download links are on the app page.

FAQ

What is the best app to identify plant diseases?

It depends on who you are and what you grow. Agrio suits advisors and commercial growers, PlantVillage+ suits field staff who also want records, and PictureThis suits home gardeners. The studies cited here do not rank them for all crops, so run the five-step check on your own crop.

Can a plant disease app be trusted without a lab test?

For a low-cost, low-risk decision, a clear result backed by a good fact sheet may be enough. For an expensive spray, an unusual symptom or a suspected quarantine disease, confirm it with a lab or advisor. The PlantVillage+ developers themselves say AI results should support expert advice.

Do these apps work without internet?

Some can. The 2020 Nuru study described diagnosis with no mobile network for the cassava model. Availability varies by crop and device, so test it in your field with data turned off.

Why does the app give a different answer for each photo?

Image models respond to light, angle, background and which leaf you choose. That is why the cassava trial gained so much from using six leaves. Take several photos, then look for the answer that comes up repeatedly.

Can a satellite identify a plant disease?

Not by name in routine use. Vegetation maps show where crops are changing, which helps you decide where to scout. You still need a look on the ground, and often a lab, to learn the cause.

Authored by: Farmonaut Research & Content Team

Reviewed and verified by: Farmonaut Agricultural Experts Team

Reviewed November 2026 against peer-reviewed studies in Frontiers in Plant Science and Nature Ecology and Evolution, university diagnostic lab pages, UK government guidance and each tool’s own store page or website.

Farmonaut crop monitoring

See where to scout first

Draw your fields once and Farmonaut shows satellite vegetation maps through the season, so you can choose where to walk and photograph.

Open The Farmonaut Crop Monitoring Web AppGet The Farmonaut App On Google PlayGet The Farmonaut App On The App Store
Farmonaut Web App With A Satellite Crop Health Map

Sources

  1. Mohanty, Hughes and Salathรฉ, Using Deep Learning for Image-Based Plant Disease Detection, Frontiers in Plant Science (2016): frontiersin.org
  2. Mrisho et al., Accuracy of a Smartphone-Based Object Detection Model, PlantVillage Nuru, Frontiers in Plant Science (2020): frontiersin.org
  3. Savary et al., The global burden of pathogens and pests on major food crops, Nature Ecology and Evolution (2019): nature.com
  4. Boulent et al., Convolutional Neural Networks for the Automatic Identification of Plant Diseases, Frontiers in Plant Science (2019): frontiersin.org
  5. Agrio, App Store listing (viewed October 2026): apps.apple.com
  6. Agrio, Google Play listing (viewed October 2026): play.google.com
  7. PlantVillage+, Google Play listing (viewed October 2026): play.google.com
  8. PictureThis, App Store listing (viewed October 2026): apps.apple.com
  9. Pl@ntNet, project website: plantnet.org
  10. Crop Protection Network, website: cropprotectionnetwork.org
  11. National Plant Diagnostic Network, website: npdn.org
  12. Virginia Tech Plant Disease Clinic, Submitting Samples: spes.vt.edu
  13. Texas Plant Disease Diagnostic Lab, Fees: plantclinic.tamu.edu
  14. University of Rhode Island, Plant Diagnostic Laboratory: web.uri.edu
  15. UK government, Plant health controls guidance: gov.uk
  16. Royal Horticultural Society, RHS Gardening Advice: rhs.org.uk


Discussion

Questions, corrections or field experience? Share them below. First-time comments are reviewed before they appear, and your email is never published.

No comments yet. Be the first to add a question or your experience.

Leave a comment

Farmonaut Farmonaut Trusted by 200,000+ users and 100+ businesses 200,000+ users trust us Get started