Non-destructive acoustic screening of pineapple ripeness by unsupervised machine learning and Wavelet Kernel methods

Area of Science:

• Agricultural Engineering
• Signal Processing
• Machine Learning

Background:

• Manual screening of pineapples for export is labor-intensive and prone to human error, leading to inconsistent quality assessment.
• Prolonged auditory inspection by human workers results in decreased concentration and arbitrary judgments, impacting efficiency.
• There is a need for an objective, automated system to ensure consistent fruit quality for long-haul transportation.

Purpose of the Study:

• To develop a non-destructive, automated device for classifying pineapple ripeness based on acoustic properties.
• To implement advanced signal processing and machine learning techniques for accurate fruit sorting.
• To overcome the limitations of manual inspection in large-scale pineapple export operations.

Main Methods:

• A novel screening device was developed that taps pineapples and analyzes the resulting sounds.
• Wavelet kernel decomposition and unsupervised machine learning (ML) were employed for sound analysis and classification.
• Acoustic couplers and a specifically designed thorn-board were utilized to optimize sound transmission and data acquisition.

Main Results:

• The developed unsupervised ML method achieved a high accuracy of 98.56%.
• An F1-score of 0.93 was obtained, indicating robust classification performance.
• The system demonstrated effective non-destructive screening of pineapples, surpassing manual inspection consistency.

Conclusions:

• The automated acoustic screening device offers a reliable and accurate solution for pineapple quality assessment.
• The integration of wavelet decomposition and unsupervised ML provides an effective approach for non-destructive fruit analysis.
• This technology has the potential to significantly improve efficiency and consistency in the fruit export industry.