Article Highlight | 19-Nov-2026

A “lightweight hunter” for tomato leaf diseases and pests: Multiscale perception and detail enhancement unlock high-precision detection

Higher Education Press

Agricultural production is directly related to global food security, and crop health has a significant impact on agricultural yield and quality. Tomato is one of the world’s important economic crops, but during cultivation it is susceptible to a range of fungal, bacterial, and viral diseases, as well as mite pests. Among these, tomato leaf diseases are particularly common and diverse, including early blight, late blight, leaf miner, leaf mold, tomato yellow leaf curl virus, Septoria leaf spot, red spider mite, and yellow curl leaf virus. Traditional detection methods based on handcrafted features and threshold segmentation have limited generalization ability in the complex environments of real agricultural production; traditional machine learning methods require a high degree of expertise in feature selection and struggle to handle high-dimensional, large-scale image data. In recent years, deep learning object detection algorithms represented by the You Only Look Once (YOLO) series have been widely used in agricultural disease detection because of their real-time performance and high computational efficiency. However, existing models still face challenges such as large variations in lesion scale, blurred boundaries, difficulty in identifying tiny lesions, and complex lighting and background interference.

Recently, the team led by Hui Liu at the School of Traffic and Transportation Engineering, Central South University, published a research paper titled “A lightweight tomato leaf disease detection method integrating multiscale perception and detail enhancement” in ENGINEERING Agriculture, an Engineering cluster journal of the Chinese Academy of Engineering (DOI: 10.15302/J-FASE-2026682). Based on an improved YOLO11n framework, the study proposed a lightweight and efficient tomato leaf disease and pest detection method.

First, a dynamically refined intersection over union (DRIoU) loss function is introduced, which uses a sigmoid-like function to dynamically adjust the IoU scaling ratio and optimize bounding box regression quality and stability across different training stages. Second, an adaptive multiscale fusion (AMSF) module is designed. It divides the input features into three groups along the channel dimension, with 50% of the channels retained directly to preserve details, while the other two groups are processed by 3×3 and 5×5 convolutions to extract medium- and large-scale lesion features, respectively, and then fused through a 1×1 convolution, enhancing feature extraction adaptability to lesions of different sizes. Third, a progressive receptive field via dilated convolutions (PRDC) module is proposed. It replaces the traditional spatial pyramid pooling fast (SPPF) module with 3×3 dilated convolutions with shared weights and dilation rates of 1, 3, and 5, progressively expanding the receptive field and improving spatial perception of lesions under complex backgrounds. Finally, a detail enhanced detection head (DEDH) is introduced. Across the P3–P5 scales, it shares a detail-enhancement convolution (DEConv) module containing five parallel branches—center-difference, horizontal-difference, vertical-difference, asymmetric-difference, and standard convolution—thereby strengthening lesion boundary and texture detail perception.

Experimental results showed that the method performed well in tomato leaf disease and pest detection. Compared with the YOLO11n baseline model, the proposed method improved mean Average Precision at an IoU threshold of 0.5 (mAP50) by 2.1% and mean Average Precision averaged over IoU thresholds from 0.5 to 0.95 (mAP50–95) by 2.9%, while increasing inference speed to 482 frames per second (FPS).

The method proposed by the research team achieves a good balance among detection accuracy, real-time inference, and lightweight deployment capability. It provides a practical and scalable solution for robust tomato leaf disease detection under real field conditions and offers valuable guidance and technical support for the future development of intelligent agricultural pest and disease monitoring.

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.