A lightweight deep learning model for detection of Huanglongbing disease

As a continuously spreading citrus disease worldwide, citrus Huanglongbing(HLB) disease causes serious economic losses to the citrus industry every year. Currently, no effective curative method is available. Therefore, the development of accurate and efficient HLB detection methods is crucial for curbing its spread and mitigating economic losses. Owing to the diverse symptomatology of HLB, traditional detection methods are often time-consuming and labor-intensive, hindering large-scale and efficient monitoring. Consequently, numerous neural network models have been developed for HLB detection in recent years. However, the performance of a given model can vary significantly when processing images acquired from different devices. To address these challenges, we propose the YOLOv5n-EMA-GAM model and evaluate its performance on images captured under diverse acquisition conditions. The results demonstrate that our model achieves superior accuracy across different acquisition conditions while exhibiting enhanced robustness (smaller performance variance). Furthermore, to enhance model performance, we propose a novel loss function to handle low-quality samples within the dataset. Experimental results indicate that compared to models employing CloU, Focal-EIoU, and Wise-IoU v3 loss functions, our model reduces the average error rate by 1.68%, 1.14%, and 2.05%, respectively, and compared with other lightweight models such as MobileNetV3, EfficientNetV2, and ShuffleNetV2, the average error rates have been reduced by 0.59%, 3.12%, and 0.5%, respectively. Moreover, our model achieves an average error rate comparable to YOLOv8n and YOLOv11n, but with fewer parameters. The number of parameters in the model is 2.37M, which is conducive to hardware deployment and enables automated detection. This work presents an efficient model for HLB detection. When integrated with other management strategies, it can significantly enhance HLB prevention and control, thereby mitigating its impact on the citrus industry.