Abstract: The size, shape, and arrangement of tea leaf serrations are important criteria for assessing and evaluating tea germplasm resources. However, for a long time, the determination of these criteria has mainly relied on the experience of technicians, resulting in subjective judgments and uncertainties. In this study, a custom convolutional operator based on the geometric morphological features of leaves was introduced, and the existing image recognition algorithms were optimized. A quantification method for tea leaf serrations based on leaf image analysis was proposed. Through experimental validation, the results show that this method can rapidly and accurately obtain basic parameters such as leaf area, perimeter, and number of serrations. It can also quantify descriptive indicators such as serration sharpness, serration depth, and serration density using custom operators. The coefficients of variation for the quantified results are all less than 1%, indicating strong repeatability and high stability. Compared to manual subjective identification, the measurement time of this method does not exceed 30 s, effectively enhancing the accuracy and efficiency of evaluating the morphological characteristics of tea leaf serrations. It provided a new approach and perspective for the quantitative evaluation of tea germplasm resources.

Key words: tea leaf, morphological quantification, serration density, serration depth, serration sharpness