Unfortunately, no Asian hornets were sighted again this month. The stations in Munich, however, were practically besieged by European hornets. Regrettably, YOLO’s classification is not as accurate as hoped, resulting in many false positives.<\/p>\n\n\n\n
Disabling automatic exposure also led to images that were either heavily overexposed or very dark. Since the European hornet is nocturnal and many events occur during the morning and evening hours, this setting was switched back to automatic after a few days. Switching to manual zoom, on the other hand, significantly improved image quality. Only the AI camera on Pi-4 has a hardware zoom, which is very difficult to adjust due to its complicated installation.<\/p>\n\n\n\n
The hot El Ni\u00f1o summer is also taking quite a toll on the stations. The adhesive on one camera module has melted, and the wooden plates are splintering at the corners.<\/p>\n\n\n\n
![\"\"](\"https:\/\/hornet-radar.com\/wp-content\/uploads\/2026\/09\/kamera-ausfall2.jpg\")
<\/figure>\n<\/div>\n<\/div>\n\n\n\nHardware Adjustments:<\/strong><\/p>\n\n\n\n The AI camera module on station Pi-4 is being replaced with a Camera Module 3, in line with the other stations, and the camera with the melted adhesive has been repaired. Additionally, the opening of the bottle containing the lure liquid has been repositioned lower so that it sits at the same height as the landing board, allowing a view of the hornets’ bodies that is as uncurved as possible.<\/p>\n\n\n\n Software Adjustments<\/strong>:<\/p>\n\n\n\n In contact with the English research team, a discrepancy in aspect ratio was identified, so frames are now captured in 16:9 format at a resolution of 1080p and downscaled by a factor of 0.75 before classification, which promises significantly better results.<\/p>\n\n\n\n Until now, the system was also only able to track a single insect per camera frame \u2014 it always picked the first moving object and ignored all others. This issue was known but had been deferred for the time being.<\/p>\n\n\n\n With the July update, the system is being rebuilt for multi-object tracking: each detected insect now receives its own ID that remains stable across frames and is tracked individually. In addition, the YOLO classification is being moved out into its own worker daemon thread and limited to 2 CPU cores, leaving enough computing power for tracking and significantly improving performance.<\/p>\n\n\n\n <\/p>\n","protected":false},"excerpt":{"rendered":" Results: Unfortunately, no Asian hornets were sighted again this month. The stations in Munich, however, were practically besieged by European hornets. Regrettably, YOLO’s classification is not as accurate as hoped, resulting in many false positives. Disabling automatic exposure also led to images that were either heavily overexposed or very dark. Since the European hornet is … <\/p>\n