Enhanced YOLOv8n for Machine Room Indicator Light Detection

Introduction: This study addresses the critical challenge of insufficient detection accuracy for small targets such as indicator lights in industrial station facilities by enhancing the YOLOv8n model. Although existing patented technologies offer various detection solutions, their recognition precision for minute indicators falls short of meeting industrial reliability requirements. Methods: First, the CMUNeXtBlock module employs large-kernel depthwise separable convolution and an inverted bottleneck design to enhance local-global feature fusion. Second, the lightweight Lighter-BiFPN architecture optimizes multi-scale feature interactions. Third, a dedicated 160 × 160 high-resolution detection branch enhances the extraction of small-target features. Finally, the WIoU loss function refines bounding box regression accuracy. Results: Experimental validation confirms significant performance gains. Compared to the baseline YOLOv8n, our improved model achieves a 3.9% increase in Precision (reaching 94.9%), a 6.2% higher Recall (96.8%), and a 3.9% improvement in mean Average Precision (mAP = 95.9%). These metrics demonstrate substantial advancements in small-target detection capability, with mAP exceeding typical patent-reported performance (≤ 91.5%) by 4.4 percentage points. Discussion: The performance enhancements validate that our integrated approach effectively resolves core challenges: CMUNeXtBlock mitigates weak feature representation, Lighter-BiFPN reduces missed detections across scales, the high-resolution branch improves small-target discrimination, and WIoU corrects positioning inaccuracies. This 40% improvement in multi-scale processing efficiency over patent solutions translates to critical industrial value for equipment monitoring systems, providing reliable visual inspection while reducing potential safety risks from detection failures. conclusion: The optimized solution successfully achieves an optimal balance between detection accuracy and computational efficiency, making it particularly well-suited for industrial implementations with stringent performance demands. Conclusion: The demonstrated improvements (Precision +3.9%, Recall +6.2%, mAP +3.9%) directly address the industry-wide pain point of insufficient small-target recognition accuracy in existing patented technologies. Our solution provides an efficient and high-precision framework for industrial applications that require mission-critical reliability in small-target detection.