Classification of Maturity Stages of Coconuts using Deep Learning on Embedded Platforms

Abstract

India stands 3rd in producing coconuts in the world, with respect to area and yield collectively contributing to sustain millions of families. These coconuts are typically harvested by climbing the trees with the use of ropes, which is a challenging task. The need to find the right coconut maturity stage is essential since different coconut stages have various benefits. Maturity detection takes the front seat in deciding the value of the coconut and is directly linked to the quality of the product. This study has observed the maturity stages of coconuts and segregated them into five classes. Further, different state of the art architectures such as Xception, ResNet50V2, ResNet152V2 and MobileNetV2 are compared to address the task of detecting the maturity stages of coconuts. Among these architectures, MobileNetV2 architecture gave the best results. MobileNetV2 was trained on the proposed dataset. It is observed that the model gives 99 % accuracy on test data. Further, the model was deployed on an Android device, making it easier for farmers to recognize different stages of coconut maturity for harvesting and other applications.