Artificial Intelligence (AI) in Agriculture — Crop Protection Reviewer Questions

170 board-style Artificial Intelligence (AI) in Agriculture items for the Agriculturist Licensure Examination, free and open to every examinee. Group pesticides by mode of action rather than by trade name. The examination asks about resistance management far more often than about products.

170 built-in questions in this topic · approved additions may publish live · part of Crop Protection

Practise Artificial Intelligence (AI) in Agriculture free

Sample Artificial Intelligence (AI) in Agriculture questions with answers and explanations

Board-style items taken from the Crop Protection bank. Every answer is explained, which is the part that makes a review question worth doing twice.

A PhilRice researcher has thousands of rice-leaf photographs already identified by experts as healthy, rice blast, bacterial leaf blight, or brown spot, and wants a computer to assign one of those names to new photographs. What does the researcher need to do?
- A. Feed in unidentified photographs and treat every unusual leaf as a new disease
- B. Let the computer invent its own disease names from unidentified photographs
- C. Give the computer only pesticide cost data and let it work out the diseases
- D. Train the model on the expert-identified photographs so it learns the visual pattern of each named category correct
Why: A recognition model learns from examples that have already been correctly identified. Because the expert labels exist and the required output is one of those named categories, the labeled photographs are the training material.
A disease classifier performs well on uniformly lit laboratory leaf images but performs poorly on farmer smartphone photos with complex backgrounds. What should the team do first?
- A. Remove difficult field images from evaluation
- B. Raise the confidence threshold until reported accuracy improves
- C. Add representative field images with varied lighting, backgrounds, cultivars, and devices, then retrain or fine-tune correct
- D. Increase the number of disease classes without adding field data
Why: The model learned on images that do not look like the ones it now receives. Only field data covering that variation will close the gap; the other options hide the weakness rather than fix it.
A corn classifier gives probabilities of 0.52 for northern corn leaf blight and 0.44 for gray leaf spot. What is the best operational interpretation?
- A. Both diseases are definitely present because both probabilities are high
- B. The 0.52 score proves northern corn leaf blight is present
- C. The result is uncertain and should trigger closer inspection or confirmatory diagnosis correct
- D. The second-highest class should always be ignored
Why: Two scores this close mean the model cannot separate the two diseases on this image. Acting on the higher one without further evidence would not be justified.
A model trained mostly on Central Luzon rice images will be deployed in Luzon, Visayas, and Mindanao. Which test design gives the strongest evidence of generalization?
- A. Randomly split individual images from the same plants across train and test
- B. Report training accuracy from the full dataset
- C. Tune the model repeatedly on the final test set
- D. Hold out farms or regions not used in training and test there correct
Why: Testing on farms and regions the model has never seen matches how it will actually be used. A random split of images from the same plants makes the test far easier than deployment.
An extension worker receives an AI pest-classification result from a farmer's phone. How should it be used in IPM?
- A. Use it to narrow the diagnosis, check confidence and image quality, and verify unusual or high-consequence cases before recommending control correct
- B. Treat the predicted label as more reliable than field scouting
- C. Ignore economic thresholds because AI already identified the pest
- D. Automatically recommend the strongest pesticide for every positive result
Why: An image result narrows the possibilities. Confirmation of the pest and the IPM decision rules -- thresholds, natural enemies, stewardship -- still govern what is recommended.
A sugarcane estate has three years of trap counts, weather records, crop stage and field locations, but no expert has rated any field as high or low risk. The manager wants to find out which fields behave alike in terms of pest pressure. Which approach is most suitable?
- A. Fit a leaf-image segmentation model to the trap and weather records
- B. Group the fields by the similarity of their records, then describe what each group has in common agronomically correct
- C. Train a model to predict risk ratings that have never been recorded
- D. Ask the system to select an insecticide before the patterns are examined
Why: With no ratings to learn from, the useful step is to let the records fall into groups of similar fields and then interpret those groups agronomically. Grouping suggests where to scout; it does not by itself predict risk.
A clustering analysis groups banana fields by humidity, leaf wetness, and disease observations. Which conclusion is justified?
- A. Each cluster is equivalent to a pathogen species
- B. The largest cluster should automatically receive fungicide
- C. The clusters prove humidity is the disease cause
- D. The clusters reveal similar multivariate patterns but do not by themselves prove that humidity caused disease correct
Why: Grouping shows that certain fields behave alike. It does not establish that humidity caused the disease, and it does not set a treatment threshold.
An extension network receives thousands of new farmer-submitted pest photos each month. How should these data be incorporated?
- A. Let farmer captions replace all expert annotations
- B. Automatically accept every model-generated label as verified ground truth
- C. Use them to improve representation, but route uncertain or novel-looking cases for expert labeling before model updates correct
- D. Delete rare images because they slow training
Why: Farmer photos are exactly the kind of data the model needs, but their identifications are unverified. Expert review of the uncertain and unusual cases keeps label quality from degrading as the collection grows.

Practise the full Crop Protection bank free

Artificial Intelligence (AI) in Agriculture — Crop Protection Reviewer Questions

Sample Artificial Intelligence (AI) in Agriculture questions with answers and explanations

A PhilRice researcher has thousands of rice-leaf photographs already identified by experts as healthy, rice blast, bacterial leaf blight, or brown spot, and wants a computer to assign one of those names to new photographs. What does the researcher need to do?

A disease classifier performs well on uniformly lit laboratory leaf images but performs poorly on farmer smartphone photos with complex backgrounds. What should the team do first?

A corn classifier gives probabilities of 0.52 for northern corn leaf blight and 0.44 for gray leaf spot. What is the best operational interpretation?

A model trained mostly on Central Luzon rice images will be deployed in Luzon, Visayas, and Mindanao. Which test design gives the strongest evidence of generalization?

An extension worker receives an AI pest-classification result from a farmer's phone. How should it be used in IPM?

A sugarcane estate has three years of trap counts, weather records, crop stage and field locations, but no expert has rated any field as high or low risk. The manager wants to find out which fields behave alike in terms of pest pressure. Which approach is most suitable?

A clustering analysis groups banana fields by humidity, leaf wetness, and disease observations. Which conclusion is justified?

An extension network receives thousands of new farmer-submitted pest photos each month. How should these data be incorporated?

Other Crop Protection topics