Aerial Images and Convolutional Neural Network for Cotton Bloom Detection

Abstract

Monitoring flower development can provide useful information for production management, estimating yield and selecting specific genotypes of crops. The main goal of this study was to develop a methodology to detect and count cotton flowers, or blooms, using color images acquired by an unmanned aerial system. The aerial images were collected from two test fields in 4 days. A convolutional neural network (CNN) was designed and trained to detect cotton blooms in raw images, and their 3D locations were calculated using the dense point cloud constructed from the aerial images with the structure from motion method. The quality of the dense point cloud was analyzed and plots with poor quality were excluded from data analysis. A constrained clustering algorithm was developed to register the same bloom detected from different images based on the 3D location of the bloom. The accuracy and incompleteness of the dense point cloud were analyzed because they affected the accuracy of the 3D location of the blooms and thus the accuracy of the bloom registration result. The constrained clustering algorithm was validated using simulated data, showing good efficiency and accuracy. The bloom count from the proposed method was comparable with the number counted manually with an error of −4 to 3 blooms for the field with a single plant per plot. However, more plots were underestimated in the field with multiple plants per plot due to hidden blooms that were not captured by the aerial images. The proposed methodology provides a high-throughput method to continuously monitor the flowering progress of cotton.

High-Throughput Field-Phenotyping Tools for Plant Breeding and Precision Agriculture

“…Tattaris et al [23] used an octocopter mounted with multispectral and thermal sensors to evaluate spring wheat trials for yield and biomass at an altitude between 30 and 100 m. Xu et al [31] used an octocopter with an RGB camera to evaluate cotton trials at 15 m flight height. A hexacopter with a thermal camera was used to evaluate drought tolerance in black poplar with flights at 25 m altitude [32].…”

“…GCPs can drastically increase the absolute accuracy of the orthophoto to 2-5 cm both horizontally and vertically. For example, Khan et al [33] by flying at a lower altitude (20 m) obtained lower GSD and further incorporated GCPs to produce accurate orthophotos; whereas Xu et al [31] used a Lumix DMC-G6 camera with a 16 megapixel sensor at an altitude of 15 m to reduce GSD. However, information was not provided on whether GCPs were also used for further rectification.…”

Pear Flower Cluster Quantification Using RGB Drone Imagery

“…The application of drones is usually preferred over a ground vehicle or robot to collect this data, as a drone can provide superior data collection speed and larger spatial coverage. In addition, drones do not interact with the plants, so constant data collection will not cause soil compaction and plant damage, which can happen when using a ground vehicle [8].…”

“…To the best of our knowledge, previous studies using proximal data for flower detection and classification, only focused on Red-Green-Blue (RGB) and multispectral sensors [8][9][10][11][12]. As flowers generally have a distinct color from the background, the three or four bands of these sensors suffice for distinguishing the flowers from the rest of the orchard scene.…”

“…The problem with this methodology is that it assumes that flowers or flower clusters always have the same size. This is not the case since they can differ in size due to occlusion by the canopy [16], the phenological stage they are in [8] and differences in scale due to the position of the flowers on the tree along the z-axis. In contrast to pixel-based methods, object-based classification methods first aggregate spectrally homogenous image objects using an image segmentation algorithm, followed by a classification of the individual objects [13].…”