Design and Development of a Neural Network — Based Coconut Maturity Detector Using Sound Signatures

Revolutionizing the post-harvest process using low-cost non-destructive approaches such as acoustics is a promising frontier to uplift the conventional farming practices as well as adapt machine learning tools to further increase the accuracy of the system. This study attempts to develop a prototype that uses the sound signatures to classify the maturity level of the young coconut by using neural network. Specifically, it aims to: (1) extract the acoustic features from the sound signal; (2) train the data set and develop the classification model using neural network; (3) develop a program for the prototype; (4) design and fabricate the hardware for the prototype; and (5) test and evaluate the protype. The major parts of the prototype were the vibration motor, vibration sensor, rpi 3b+ microcontroller, battery and LCD. Results of the neural network model obtained an overall classification accuracy of 91.3 % and an R-value of 0.94229. This implied that the neural network model has a high prediction rate to accurately determine the maturity level of the coconut using the sound signatures. Lastly, final evaluation results showed that the prototype has a higher percentage of prediction accuracy as compared to the manual process.