The study, carried out by Lin Ding Shan, examines a complete disease-detection pipeline deployed on a Raspberry Pi 5, a low-cost single-board computer carried by a worker on a walking inspection round of a few hours. The choice of operational form matters. Mature durian trees reach 15 to 30 metres, so a fixed ground camera would see only trunks and low foliage; the canopy where foliar symptoms appear is out of reach. Instead, the operator supplies mobility and framing while the detector supplies diagnostic judgement. That design makes three questions decisive: whether the reported accuracy is inflated by faults in the dataset, whether a class scoring zero failed because of the model or the annotation, and whether the device can sustain inference for a whole round without suspending itself to cool down. Each failure, the paper argues, costs the grower something concrete.

The first finding concerns data leakage, and it is striking. The author subjected the dataset of 995 training and 126 validation images to a formal integrity audit using exact MD5 hashing and perceptual hashing to detect near-identical images across the train/validation boundary. The audit found one byte-identical image and 117 near-duplicate pairs spanning the split, tracing to an augmentation-before-split workflow: augmented variants of source photographs were generated first, and the enlarged pool was then divided, placing siblings on both sides of the boundary. The remedy was deliberately narrow. Holding the trained model and weights fixed, the 32 implicated validation images were removed and the evaluation repeated. The Algal_leave class fell from an average precision of 0.806 to 0.597, a 35 percent relative inflation, and aggregate mAP@0.5 dropped from 0.402 to 0.367. The audit did not merely revise a figure; it withdrew a finding, halving an apparent annotation-protocol effect that the leaked data had manufactured.

Instance-to-image tracing exposed a second, subtler problem. The 54 validation instances of Phomopsis traced to a single photograph, the 26 Leaf_rot instances to five images from one 39-second capture session, and root_disease to a single instance on a single image. Eighty-five percent of the Leaf_rot validation evidence derived from six seconds of near-continuous footage of one scene. Grouping all 1,121 exported images into source components by filename, hash identity and perceptual similarity yielded roughly 750 distinct source photographs, meaning the effective sample size was about a third smaller than the nominal image count. In such a dataset, a per-class instance count is not a measure of diversity, and any per-class score built on it is statistically fragile.

The second investigation dismantled the original explanation for rare-class suppression. The initial study had attributed near-total failure on rare classes to insufficient model capacity in the small YOLOv8n detector. To test that, the author retrained three architecture families under a single controlled environment: YOLOv8s across five random seeds, YOLOv11s and the transformer-based RT-DETR-l across three each. The result was unambiguous. Leaf_rot scored exactly 0.000 with zero variance in every one of the eight YOLO runs, and Phomopsis never exceeded 0.005, even though RT-DETR-l, with three times the parameter count of YOLOv8s and the highest aggregate score of any architecture tested, was in play. Meanwhile the difference between architectures, 0.017 in mAP@0.5, sat far below the seed-to-seed variation of 0.118 within a single architecture. Architecture moved the aggregate within its own noise band without touching the classes the aggregate was meant to diagnose.

The retraining on a corrected, source-level leak-free partition then revealed why. Leaf_rot, dead at zero on the original partition, returned 0.399 on the re-partitioned one with the same architecture and configuration. The label existed and the signal was learnable; the zero was a property of the validation sample. The author proposes a candidate diagnostic category for this, an evaluation-space failure, alongside label-space failure, where a class was never assigned a label at all, and feature-space failure at threshold. Anthracnose exemplifies the first: it carries zero validation instances because it was absorbed into the visually similar Algal_leave during annotation. Phomopsis exemplifies the second: lowering the confidence threshold recovered a detection on genuinely symptomatic tissue, showing some signal was present. The three modes demand opposite remedies, and a capacity-only diagnosis directs engineering effort at the wrong problem.

The third investigation is the one most likely to surprise edge-computing practitioners. Sustained inference on a Raspberry Pi drives the processor die toward its thermal limits, and the platform responds at two distinct levels: hardware frequency throttling, which reduces the clock, and a software thermal cut-off, which suspends inference outright at 82 degrees Celsius. The literature has largely conflated them. Across five duty-cycle configurations, each run for roughly three hours and replicated four weeks later, the study separated them decisively. Continuous inference triggered 99 protective cut-offs in three hours, roughly once a minute; a 15-second sleep interval still triggered 41; a 30-second interval triggered none. Yet throttling itself cost only 2.6 percent of median inference latency, 419 milliseconds under continuous operation against 408 milliseconds under a schedule that never throttled at all. The mechanism the field conventionally optimises is nearly costless here; the one it ignores is what ends the deployment.

The energy measurements added a counterintuitive twist. Mean power fell 28.2 percent from continuous operation to the longest sleep interval, but energy per inference rose 39.4 percent, from 3.27 to 4.56 joules, because the roughly 2.2-watt idle floor is paid throughout every sleep window while producing nothing. Continuous operation is the most energy-efficient way to perform a fixed number of inferences, and it is also the configuration that cannot sustain them. A 30-second sleep interval costs 23.7 percent more energy per detection and removes the cut-offs. The author’s design criterion follows directly: choose the shortest sleep interval for which the cut-off count is zero, and accept the throttling that remains. Under the tested indoor conditions that interval is 30 seconds, delivering 66.7 percent effective monitoring coverage, though a warmer third round showed the margin is thin and outdoor deployment should treat these figures as lower bounds.

Translated to the orchard, the numbers acquire human weight. In a recorded 2.25-hour field session, the device spent 91.2 percent of its time with the throttling flag set and peaked at 81.5 degrees Celsius, and 27 inferred cut-off events cost 135 seconds of suspended monitoring, about 1.66 percent of the round. At a walking speed of 1.2 metres per second along rows spaced 10 metres apart, that amounts to roughly 16 trees of about 970 encountered being walked past unexamined, with no signal indicating when. The author suggests a cheaper intervention than hardware changes: surface the suspension state to the operator, so a worker can hold position for the five seconds a cut-off costs rather than continue past a tree the node is not examining. The dataset lesson is equally operational. A grower shown the uncorrected 0.806 would calibrate trust against a performance level the system does not attain and under-inspect precisely the classes it reports most confidently.

The study is candid about its limits: one dataset, one device, one software stack, indoor thermal characterisation with a single outdoor session, no independent test partition, and no field accuracy evidence at all. But its methodological prescriptions travel well beyond durian. Split before augmenting; audit the train/validation boundary with exact and perceptual hashing before reporting per-class metrics; report instance counts alongside the number of distinct source photographs behind them; and distinguish protective suspension from frequency throttling when characterising an edge device. The deeper message is that aggregate metrics conceal exactly the failures that decide whether a deployment works, whether that failure is a leaked image inflating a score, a class silenced by its validation sample, or a processor quietly stopping to cool while trees go uninspected. For agricultural AI moving from benchmark to field, the quantity to design against is not the one that slows the device down. It is the one that stops it.

Subject of Research: Edge-deployed deep learning disease detection in durian, covering dataset integrity auditing, rare-class failure diagnosis and thermal duty-cycle scheduling on a Raspberry Pi 5

Article Title: Thermal cut-off, not frequency throttling: Duty-cycle scheduling and a dataset-integrity audit for on-farm edge disease detection in durian

Article References: Shan, L. D. (2026). Thermal cut-off, not frequency throttling: Duty-cycle scheduling and a dataset-integrity audit for on-farm edge disease detection in durian. Smart Agricultural Technology, 15, Article 102526. https://doi.org/10.1016/j.atech.2026.102526

Image Credits: AI Generated

DOI: 10.1016/j.atech.2026.102526

Keywords: durian, edge computing, Raspberry Pi 5, plant disease detection, data leakage, perceptual hashing, thermal throttling, duty-cycle scheduling, YOLO, RT-DETR, Phytophthora palmivora, precision agriculture

The scale of the problem the researchers document is stark. In 2023, estimates suggested that around 500,000 AI-generated non-consensual intimate videos existed online. By 2025, that figure had grown to an estimated 8 million, with the overwhelming majority of targets being women and girls. The ease with which such content can now be generated from a single photograph, and then distributed through websites, social media platforms and private messaging services, many of them protected by end-to-end encryption, has allowed the volume of material to multiply at a pace that far outstrips current responses.

The harms associated with this form of digital abuse are increasingly well documented. Sarah Morgan noted that anyone with a single photo available online can be quickly and easily targeted in content that falsely depicts them in sexual or intimate contexts of any nature, without their consent, and that emerging evidence demonstrates the wide-ranging and persistent harms of this abuse. A 2024 investigation by Channel 4 News found that almost four thousand famous people were listed across five of the most visited deepfake pornography websites, including female actors, journalists, musicians, television stars and YouTubers whose faces had been superimposed onto pornographic material using artificial intelligence. Industry analysis cited in the study further indicated that 40 of the most visited deepfake pornography sites had accumulated more than four billion views.

To understand how these sites survive online, the research team examined the technical architecture that underpins them. Infrastructure, in this context, refers to the technologies and systems, such as website hosting, domain name servers and content delivery networks, that enable a website to deliver content to users. Without these services, a site cannot successfully function, no matter where in the world its operators are based. This dependency creates a potential point of leverage: if infrastructure providers refuse to serve sites hosting abusive content, those sites become dramatically harder to reach.

The study is the first systematic attempt to identify and name the key players, whether knowingly or unknowingly, facilitating the technological infrastructure supporting sites that host AI-generated non-consensual intimate imagery. The researchers identified 88 sites hosting such content, of which 38 exclusively hosted it. In capturing contextual data relating to each site, they found that some hosted hundreds of thousands of videos. The content primarily targeted Hollywood and Hindi-film stars, K-pop idols, and well-known feminists and activists, with nearly all of the material featuring females.

Among the findings, five key players emerged as the dominant infrastructure providers enabling the sites to function, with Cloudflare identified as the most recurring name. The team also assessed how easily the sites could be found through ordinary web searches, revealing that 93 percent of them were readily discoverable via Google Search. This discoverability matters because it means that victims, and indeed anyone curious, can stumble upon the abusive material with minimal effort, amplifying the reach and impact of the original violation.

Legislative responses are beginning to emerge in several jurisdictions. England and Wales, among other countries, have moved to criminalise the sharing, creating or soliciting of AI-generated non-consensual intimate imagery. However, Morgan argues that legislation alone is not enough. Such laws will offer some deterrence, she said, but further steps are needed to prevent the abuse from occurring in the first place. She posed the question of whether the distribution supply can be cut, arguing that this abuse needs to be prevented and that putting pressure on the technology platforms which distribute such content to the public may be the most optimal chance of reducing what she described as the wildfire spread of AI-generated sexual abuse.

The researchers issued specific technical recommendations for infrastructure providers. They call on the companies facilitating AI-NCII-hosting sites to take a firm stance on the abusive content: providers should block sites that exclusively host AI-generated non-consensual intimate imagery, and suspend services for sites that host such content within a larger offering until good-faith efforts are made to remove the abuse. They also recommend deploying preventative safeguards, such as utilising perceptual hashing to extract digital signatures of confirmed abusive content and stop its redistribution. Perceptual hashing is a technique that generates a fingerprint of an image or video based on its visual characteristics, allowing identical or near-identical copies to be detected and blocked even if file names or metadata are changed, a method already used against child sexual abuse material and terrorist content on major platforms.

Dr Sophie Nightingale added that it is also time to stop blaming the targets of this abuse. All too often, she said, the victims of abuse that is sexual in nature are attributed responsibility, and there is an urgent need to change this. Blame, she argued, should be assigned to the perpetrators who create, distribute and facilitate the sharing of fake content that has the potential to ruin lives. This reframing shifts the focus from questions about what victims did or did not share online to the conduct of those who manufacture the imagery and the companies whose services make its distribution possible.

The Lancaster University team is continuing to examine the impacts and harms of non-consensual intimate imagery more broadly, and has opened a confidential survey inviting people to share their views on how this abuse could be tackled or to describe their own experiences. As the volume of AI-generated sexual abuse material continues to climb, the study’s central message is that the responsibility cannot rest with victims or with criminal law alone. The companies that provide the hosting, domain and delivery services forming the backbone of the web now face a direct, evidence-based call to action: identify the abuse on your network, cut off the sites built on it, and deploy the technical tools that already exist to stop the spread.

Subject of Research: Infrastructure providers enabling AI-generated non-consensual intimate imagery websites

Article Title: Call for technology providers to take action on AI non-consensual ‘deepfake pornography’

Article References: Call for technology providers to take action on AI non-consensual ‘deepfake pornography’. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: deepfake pornography, AI-NCII, Cloudflare, infrastructure, perceptual hashing, Lancaster University, Journal of Online Trust and Safety, content delivery networks, non-consensual intimate imagery, online abuse, digital harms, platform regulation