In consideration of the minimal differences in sensitivity to target features among various sensors including visual, laser, and optical fiber sensors, especially in complex backgrounds where targets exhibit significant scale invariance and subtle feature differences, this increases the challenge of target recognition in later stages using multi-sensor data. Therefore, this study proposes a multi-sensor key feature point target recognition technology based on an improved YOLOv5 algorithm. The technology aligns key points captured from target images by multiple sensors to enhance sensitivity to target features. The YOLOv5 model is divided into four parts: the input layer Input, the backbone network Backbone, the neck network Neck, and the head network Head. In the initial Focus module of the backbone network, multi-sensor data features are fused through slicing and splicing operations. For sensor data with different dynamic sensitivity ranges, normalization is first performed to adjust the values to a similar range, and then the convolutional layer parameters of the Focus module are set to the sum of the number of channels of the multi-sensor. On the basis of the Feature Pyramid Networks FPN, the C3STR structure is introduced, combined with the Cross Stage Partial Network CSP and the multi-head attention mechanism of the Transformer, to distinguish the sensitivity of subtle difference features and complete recognition optimization. Experimental results show that when using the improved YOLOv5 model for target recognition, the information entropy after image fusion reaches 64 bits, demonstrating high recognition accuracy and good effect.