A Lightweight Network With Efficient Channel Attention for Intelligent Daylily Maturity Detection Method

To tackle the problem of reduced effectiveness in manually identifying the maturity of daylily due to the shading effects of daylily on one another, a lightweight attention mechanism method called YOLOv7-LEB (lightweight efficient channel attention bidirectional feature pyramid network) is proposed. The enhanced YOLOv7-LEB algorithm introduces ghost lightweight convolution, reducing redundancy in model feature extraction. To improve detection accuracy and sensitivity, the efficient channel attention mechanism is utilized. In addition, the bidirectional feature pyramid network is employed for feature fusion, replacing the traditional concatenate approach, which enables better multichannel information integration and significantly boosts model performance. Experimental results demonstrate that the improved YOLOv7-LEB algorithm minimizes the model's weight by lowering the number of parameters and floating points by about 2.55 million and 8.1 billion, respectively. The model size has diminished from 74.8M to 69.7M, while the training time has been reduced by around 0.279 h. In addition, in terms of model performance, the average precision has increased from 91.3% to 91.7%, while [email protected] has risen by 0.384% . Moreover, the real-time detection capability reaches 100 frames/s. It can be observed that the YOLOv7-LEB algorithm significantly enhances detection accuracy and reduces model complexity. This improvement effectively advances the automation and intelligence of daylily harvesting, thereby enhancing harvesting efficiency, lowering labor costs, and ultimately achieving sustainable development in the daylily industry.