Construction and testing of mango fruit shape identification platform based on 3D point cloud data

The manual measurement of mango fruit dimensions presents several challenges, primarily characterized by time inefficiency and low measurement accuracy. Specifically, the reliance on subjective assessments and simplistic definitions of fruit shape can lead to erroneous classifications of mango morphology, thereby adversely impacting subsequent classification and packaging processes. Consequently, the establishment of a dedicated platform for the determination of morphological indicators and accurate identification of mango fruit shapes is imperative. Such a platform would streamline the sizing process and enhance shape assessment, ultimately facilitating improved handling and marketability of the fruit. Currently, the majority of methodologies employed for assessing mango fruit morphology rely heavily on human judgment, which inherently lacks objectivity. In contrast, advanced techniques such as three-dimensional (3D) scanning have been successfully applied to the morphological analysis of other fruit types, demonstrating their potential for enhancing measurement accuracy. Additionally, the integration of machine learning techniques can significantly improve the classification of fruit shapes, effectively mitigating errors associated with human assessment. In this study, we implemented a 3D scanning platform to swiftly capture the geometric data of mango fruits, thereby constructing accurate 3D models that facilitate detailed measurement of morphological indices and shape identification. We focused on 75 different mango varieties, with three replicates for each variety, measuring 10 quantitative parameters, including volume, longitudinal and transverse diameters, and fruit shape indices through 3D scanning technology. The results from the 3D scanning measurements were then compared with those obtained from traditional manual methods. The findings indicated that the precision of measurements derived from 3D scanning surpassed that of manual measurements, thus showcasing the superiority of the former. Subsequent analyses, including eigenvalue analysis, principal component analysis, correlation analysis, and cluster analysis, were conducted to categorize the mango varieties based on their distinct fruit shapes. This classification elucidated the rich genetic diversity and variability inherent within mango germplasm resources. Notably, the study affirmed that the fruit shape index, along with longitudinal and transverse diameter measurements, serve as critical indicators for a comprehensive evaluation of fruit morphology. This insight provides a robust foundation for the subsequent application of machine learning algorithms to optimize the selection of input parameters for predictive modeling. The study culminated in the development of a back-propagation (BP) neural network classification model using the SPSSPRO online data analysis platform. The model architecture consisted of 100 hidden layers, with a learning rate set at 0.1 and a total of 1000 iterations. Upon training with 157 samples, the model demonstrated an impressive success rate, achieving 94.12% accuracy when tested on 68 samples. Moreover, the training set exhibited a precision and recall rate of 0.941, with a remarkably quick recognition time of only 0.451 seconds. In conclusion, our research utilizes 3D point cloud data to provide rapid, precise, and lossless morphological information for mango fruits. The comprehensive 3D point cloud maps generated include enriched data such as fruit color information, further enhancing the overall assessment accuracy. Compared to conventional measurement techniques, this innovative approach offers numerous advantages: rapid sampling, high accuracy, reduced susceptibility to external factors, and a non-contact measurement method. This study not only advances the field of fruit morphology analysis but also establishes a significant framework for future research and application in fruit classification and assessment.