Context. Tea leaf blight (TLB) stands as one of the most destructive diseases affecting tea plants, posing a significant threat to both the yield and quality of tea crops.

Aims. Our aim is to employ efficient deep learning techniques to achieve precise remote sensing monitoring of TLB in natural environments.

Methods. We present an innovative methodology that leverages the combined power of ECDet and MobileNetv3 for the detection and severity assessment of TLB from unmanned aerial vehicle (UAV) remote sensing images. ECDet is constructed with a lightweight backbone to reduce the complexity of the model, and a MicroEA-FPN feature pyramid structure and a decoupled spatial attention-weighted head to achieve balance between focusing on the detailed information of tea leaves and extracting semantic information from small targets. In addition, transfer learning has been implemented to address the performance degradation owing to low UAV image resolution, and the MobileNetv3 is used to improve the accuracy of severity assessment.

Key results. The accuracy of our method was 78.46% in detecting TLB and 83.57% in assessing the severity levels of TLB leaves.

Conclusions. Compared with other object detection and assessing methods, this proposed method achieved a good balance by maintaining high accuracy while requiring fewer parameters and computational resources.

Implications. The proposed method will aid farmers, policymakers, and researchers in better understanding the impact of the TLB disease on tea yield and in taking timely and effective measures.