Abstract:
Sri Lanka is a luxuriant tropical land with the potential for the cultivation and hence
agriculture is considered as one of the best prospect sectors of the country. To maximize the
yield from the crops, a proper classification of harvest which aids in determining the storage
conditions and the export quality is essential. Deep learning technologies facilitate crop
recognition by enabling a computer to automatically detect a crop and determine its ripeness
level. This study introduces a real-time image processing algorithm utilizing Convolutional
Neural Networks (CNNs) to identify the maturity stages of scotch bonnet peppers. The
algorithm is designed to classify the scotch bonnet peppers into three maturity stages as
unripe, moderately ripe, and ripe, by training the CNN aid of dataset of labelled images of
scotch bonnet peppers at different maturity stages. Training the CNN through
backpropagation minimizes categorical cross-entropy loss, resulting in a testing accuracy of
89.04% and training accuracy of 91.6%. These results underscore the algorithm's real-time
effectiveness in discerning the maturity stage of scotch bonnet peppers. For scotch bonnet
peppers, the algorithm holds significant potential to substantially reduce postharvest losses
and cut production costs tied to exporting top-quality produce. Precisely discerning the
maturity stages of scotch bonnet peppers ensures the delivery of high-quality products to
consumers, concurrently optimizing storage conditions and export quality. The real-time
image processing algorithm, developed using CNNs and Python, proves to be an efficient
approach for detecting the maturity stage of scotch bonnet peppers and the approach can be
extended to diverse crops, establishing its versatility in the agricultural sector.
