Non-Destructive Classification of Avocado Fruit’s Harvest Maturity from Different Harvest Batches Using Machine Learning by Integrating Lenticel Features

This study addresses the problem of determining the avocado fruit’s harvest maturity from different harvest batches by proposing a novel image processing method in extracting peel features for machine learning-based non-destructive classification. Avocado fruit’s climacteric physiological characteristics mean that the physiological maturity of the fruit before harvest directly affects the quality of post-ripening and the ease of postharvest management. Therefore, accurately assessing the harvest maturity of the fruit will help improve the quality and the price of avocados in the market. Conventional harvest maturity assessment of avocados often relies on the farmer’s subjective experience or on single-parameter measurements. In contrast, a detailed investigation of avocado peel lenticels is conducted in this study with raw peel data collection and annotation, followed by feature extraction using image processing. After obtaining a comprehensive set of lenticel features, this study further evaluates the performance of multiple classification models, such as Multilayer Perceptron (MLP) and Support Vector Machines (SVMs), for automated maturity determination. Through rigorous experimental design and quantifiable metrics, the performances of SVM and MLP models are compared in terms of their performance on integrated features from lenticel distribution. Experimental results demonstrate that this approach significantly improves the accuracy and reliability of avocado’s harvest maturity classification from different harvest batches, further confirming the potential of lenticel features integration in avocado fruit quality assessment. Future research will focus on model optimization and the development of real-time embedded applications, with the aim of extending its utility to a broader scope of in situ intelligent avocado harvest maturity inspection.