Detection of Coconut Inflorescence Using YOLOv4 Variants, Single Shot Multibox Detector, and Aggregate Channel Features

Philippines is one of the top producers and exporters of coconut sugar around the world. Coconut sugar is a healthy and low glycemic index alternative sweetener produced from freshly harvested coconut sap. The quality of the sap as the raw material is an important determinant of coconut sugar quality and can be obtained by tapping unopened mature coconut inflorescence or spadix. Thus, early detection of coconut spadix is necessary for improving the yield of harvested coconut sap for coconut sugar production. Machine vision and deep learning through object detection is presented in this study for effective detection of coconut inflorescence. You Only Look Once v4-Cross Stage Partial connection (YOLOv4-CSP) algorithm with Darknet53 convolutional neural network model achieved the highest average precision value of 0.84 compared to other object detection algorithm including You Only Look Once v4- tiny (YOLOv4-tiny), Single Shot Multibox Detector (SSD), and Aggregate Channel Features (ACF). YOLOv4-CSP Darknet53 efficiently detected coconut inflorescences from images with environmental challenges such as illumination variations and occlusion conditions. Results of this study can be used to improve and maximize the harvesting of fresh coconut sap through early detection of coconut inflorescences.