Humans still crucial as AI image analysis for crop pest and disease management has a long way to go

Figure 1: Scarab Solutions was able to use AI image analysis to identify a large percentage of spider mites on a leaf but the results were insufficient for routine scouting efforts. In a greenhouse setting, the scouting team could move around more freely and inspect the situation more accurately and efficiently than the AI tools.

Effective pest and disease management is not only time consuming, as many know, it often comes at a price. According to the Food and Agriculture Organization of the United Nations, pests such as thrips and leaf miners along with diseases such as blights and mildews cost around $220 billion to the global economy annually – equating to between 20% and 40% of annual global crop production.

To improve horticultural industry efficiency, crop management strategies will inevitably require improved techniques and technology. With technological advances in scouting already set to transform the sector, is artificial intelligence (AI) the answer?

In early development, AI promises to make a difference, but will this happen in practice? Here’s one agriculturalist’s opinion on why growers shouldn’t be too quick to completely replace their existing processes with drones and robots.

When AI meets horticulture – a promising start

As AI’s potential grows, developers set their sights on horticulture with the belief AI-driven image analysis can automate crop management operations – but are we at this stage already?

Recent developments include a ‘robot scout’ equipped with near-infrared image cameras to detect powdery mildew and image analysis to predict bud and flower yields, and the IRIS Scout Robot. With the latest remote pest monitoring system that uses machine learning (ML), growers can receive pheromone trap image analysis – crucial to real time intervention strategies.

More wide-spread cases promote using smartphone applications to scan photos for pests and diseases, often presented as or nearly ready for prime time use. So, growers using smartphone image analysis for crop pest and diseases identification cannot be far off?

A rosy picture at odds with reality

As it stands, studies suggest image analysis falls short of its promises. According to a recent Scientific American article, the effectiveness of image analysis statistics is often misleading at best. The most common ‘pairing test’, testing the ability to compare two plant images and state which has the pest or disease, gives more accuracy than multiple image analysis with no information on whether any have the pest or disease.

 … with the added challenge of false positives

The false positive issue is all the evidence needed to caution against using inaccurate or skewed AI generated results as a basis for potentially damaging pesticide use.

Consider an imaging system giving a five-percent false positive reading for blight – a conservative figure, even by current app accuracy claims. In a full field of blight, this would not pose an issue, but in a field with zero blight occurrences, a slightly different story emerges. If you take 2,000 images in that field – the number of observation points a skilled scout manages per day – you would yield 100 positive results!

You must then decide whether to act or inspect the “positive” locations to verify whether they are indeed affected. Multiply this by other pests and diseases the image analysis system is checking for, with perhaps an even higher false-positive rate, and the results speak for themselves. The higher the number of false positives, the more resources are required to verify the results and all automation gains are lost.

Figure 2: AI image analysis can generate inaccurate or skewed results such as false positives, which causes growers to use more resources such as pesticide. This can create more damage than good and significantly reverse the gains of automation.

Man vs machine – an unequal comparison

The context should not be lost. Studies comparing scenarios where there is AI or no crop scouting technology at all, do not paint a realistic picture because in some cases, a system already exists which effectively records and analyzes scout-collected data. In a greenhouse setting, a scout can move their head, turn leaves, and use a magnifying glass – giving a significantly better view of the issue than smartphone image analysis ever could.

Upskill your team with the support of mobile technologies

Now is not the time to replace scouts with AI. Instead, growers should focus on supporting and enhancing human capabilities. The right digital tools should help scouts work more accurately, faster and to a greater result – not ignore their expertise.

Smartphones will continue to be key – but not primarily as an AI tool. A more realistic and proven application is data collection and mapping. Crop protection managers should empower scouts to use their inspection skills and record results as they go – building a wealth of accurate data for comparison.

This is where effective training makes a difference. Correct identification and scoring of pests and diseases, thorough sampling protocol knowledge and techniques to expedite the process harmonize the performance and accuracy of scouts’ farm-wide and are key for success.

AI may guide scouts towards correctly identifying unknown pests or diseases, but most crop scouting is about tracking the dynamic distribution of a well-known set of pests and diseases.

Working together, not apart – digital technologies can offer new insights

By combining scout-recorded data with geographical information, the results create datasets which provide an audit trail for traceability and visualization options, such as digital maps, charts, and graphs. These prove invaluable for easy identification of unique and recurring problems and patterns with few, if any, false positives.

Digital mapping combines pest and disease scouting with human expertise to optimize outcomes. Scarab Solutions sees this daily as clients use Scarab Precision crop pest and disease scouting and mapping solutions to provide a solid basis to pinpoint infestation hotspots, determine correct pesticide use, and reduce crop losses through enhanced farm management.

As datasets grow, crop protection managers can benchmark their progress against regional figures, using anonymized data from other farms.

AI can wait – humans are still the most vital commodity

An industry talking point, AI-driven image analysis still has a long way to go. For now, GPS-tracking, mobile data collection and interpretation tools are the most effective and lucrative technological solutions for crop pest and disease management. There is no doubt AI can be a real asset to extend human intelligence and enable more efficient task completion – but let’s save AI-driven image analysis on drones and robots for another time.

Dr Mikkel Grum is Research and Development Director at Scarab Solutions