Convolutional Neural Network for Cotton Yield Estimation

Mehmet Suleyman UNLUTURK, Murat KOMESLI, Asli KECELI

Abstract

The objective of this paper was to estimate the cotton yield potential of different cotton varieties using high-resolution field images based on a convolutional neural network (CNN). The yield estimation for different cotton varieties in grams in breeding studies has a great importance for the determination of superior cultivars to be commercialized. Due to the cost and excessive time consumption typical of traditional methods, alternative ways for cotton yield estimation have been investigated over the years. This paper proposes an automated system for cotton yield prediction based on color images obtained by an unmanned aerial vehicle (UAV). Two replicational field experiments including three different cotton genotypes were conducted at May Seed R&D station in Torbali, Izmir, Turkey. Three different planting patterns including three, four and six rows, respectively in ten-meter wide areas were used as experimental plots. The ground-truth yield values for a total of six hundred planted areas were obtained by weighing the harvested cotton bolls after field images were taken. Achieving an absolute difference of no more than 350 grams for 114 out of 120 planted areas which were randomly selected only for testing purposes indicates that the CNN can effectively capture important features related to cotton yield from the field images obtained by the UAV. The combination of drone technology with reliable CNN models holds great potential for optimizing agricultural practices, improving agricultural productivity, and reducing operational costs.

Keywords

Deep learning, Image processing, Convolutional neural networks, Backpropagation neural networks.

View full article