Catching a Farm Fire Early: What AI Smoke Detection Does That a Smoke Alarm Cannot
A domestic smoke alarm in a hay barn is useless, and most people who have tried one know it within a fortnight. It is not the alarm's fault. Point detectors are designed for a sealed room with still air, where smoke accumulates at the ceiling and reaches a sensor within a couple of metres. A farm building is the opposite: enormous volume, open sides, constant draught, dust that triggers false alarms until somebody takes the battery out, and nobody within earshot at two in the morning anyway. So most farm buildings have no fire detection. Not because the risk is low, but because the obvious product does not fit the building. Last updated: November 2026. Where farm fires actually start Worth being specific, because it determines where detection belongs. Hay and straw self-heating. Bales stacked above about twenty per cent moisture generate heat biologically, and a stack can reach ignition over days or weeks. This is the one that smoulders quietly before it does anything dramatic, which makes it the one most worth detecting early. Electrical faults in outbuildings. Rodent-chewed cable, an overloaded extension running a heater, a failing bearing on a fan. Unheated, unvisited buildings are where this goes unnoticed. Grain dryers and driers. Hot work in an environment full of combustible dust. The hazard is well understood and the detection usually is not. Battery charging. Increasingly the one that catches people out. Tool batteries, e-bikes, telehandler chargers, left on overnight in a corner of a workshop. Machinery parked hot. A combine or a telehandler put away with a hot manifold against chaff. The classic harvest fire, covered in our harvest security guide. Four of those five produce visible smoke well before flame. That is the opening. How camera-based detection works Rather than waiting for smoke to reach a sensor, the camera watches for what smoke and flame look like, and makes that judgement continuously across the whole space it can see. Practically that means: It sees the whole building, not one point. A camera in the apex of a Dutch barn covers the stack. A point detector covers the air immediately around itself. In a building with fifteen metres of headroom and open sides, that difference decides whether anything is detected at all. It works outdoors. Stacks in the open, fuel bowsers, machinery yards. No point detector works outside. A camera does. It tells you what it saw. An alarm tells you there is an alarm. A camera sends the clip, and you can tell in five seconds whether it is smoke from the stack or a tractor exhaust drifting past the lens. That distinction is what stops you ignoring it. It does not need the building to be sealed, heated or occupied. Which describes almost every building this matters in. Our AI fire and smoke detection cameras run the detection on the camera rather than in a data centre, which matters for two reasons: it still works when the connection drops, and it is not sending continuous video anywhere just to be watched. Being honest about the limits This is a genuinely useful technology that is routinely oversold, so here is where it falls down. It cannot see what it cannot see. Smoke inside a sealed grain bin, inside a baled stack, or behind a partition is invisible to a lens. Camera detection finds smoke once it is in the air in the camera's view. For inside a stack, a probe thermometer is still the tool, and walking the stack with one in the first six weeks after baling is still the single most effective thing you can do. Early smouldering is slow and faint. A stack heating over a fortnight may produce very little visible smoke until quite late. The camera will usually beat a point detector and a passer-by. It will not beat a temperature probe pushed into the stack. Dust, steam and low sun cause false positives. Less than you would fear with a modern model, but not zero. A camera pointed into a December sunrise over a wet yard will occasionally decide it has seen something. You want to see the clip, not just the alert. It does not fight the fire. Obvious, but worth saying. Twenty minutes of warning is only worth something if there is a plan for those twenty minutes. What twenty minutes is actually worth That is the real argument, and it is not about saving the building. A hay barn that is properly alight is lost. Nobody is saving it. What early detection buys you is the chance to: Get stock out of the adjoining building before the roof goes. Move machinery out of the path. A telehandler moved twenty metres is worth more than the barn. Get the brigade called early, with an exact location. Rural response times are what they are, and ten minutes earlier on the call is ten minutes earlier on the ground. Separate the stack. If it is caught genuinely early, pulling bales off the heating face with a telehandler can stop it. This only ever works with warning. Tell them where the water is. The most useful thing a farmer can do when the brigade arrives is point at the nearest usable supply. Where to put them A sensible starting point for a mixed holding: The hay or straw barn, in the apex, looking down the length of the stack. This is the highest-value single camera on most farms. The main workshop, covering the bench and the charging area. Battery fires start where the chargers are. The grain store or dryer, looking at the dryer itself rather than the whole shed. The machinery shed, particularly if machines come in hot during harvest. Most of these buildings have no power where the camera needs to be and no network worth the name. That is the usual practical obstacle, and it is solvable: solar power for off-grid cameras covers running one through a British winter, and HaLow across the whole farm covers getting the signal back to the house without a SIM in every building. What it costs against what it is protecting A camera on the stack is a few hundred pounds. A stack of winter fodder is tens of thousands, and the building it is in is more. The arithmetic…