Automatic Disease Detection from Strawberry Leaf Based on Improved YOLOv8

Abstract

Strawberries are susceptible to various diseases during their growth, and leaves may show signs of diseases as a response. Given that these diseases generate yield loss and compromise the quality of strawberries, timely detection is imperative. To automatically identify diseases in strawberry leaves, a KTD-YOLOv8 model is introduced to enhance both accuracy and speed. The KernelWarehouse convolution is employed to replace the traditional component in the backbone of the YOLOv8 to reduce the computational complexity. In addition, the Triplet Attention mechanism is added to fully extract and fuse multi-scale features. Furthermore, a parameter-sharing diverse branch block (DBB) sharing head is constructed to improve the model’s target processing ability at different spatial scales and increase its accuracy without adding too much calculation. The experimental results show that, compared with the original YOLOv8, the proposed KTD-YOLOv8 increases the average accuracy by 2.8% and reduces the floating-point calculation by 38.5%. It provides a new option to guide the intelligent plant monitoring system and precision pesticide spraying system during the growth of strawberry plants.