Standardizing and Centralizing Datasets to Enable Efficient Training of Agricultural Deep Learning Models

Abstract

In recent years, deep learning models have become the standard for agricultural computer vision. Such models are typically fine-tuned to agricultural tasks using model weights that were originally fit to more general, non-agricultural datasets. This lack of agriculture-specific fine-tuning potentially increases training time and resource use, and decreases model performance, leading an overall decrease in data efficiency. To overcome this limitation, we collect a wide range of existing public datasets for three distinct tasks, standardize them, and construct standard training and evaluation pipelines, providing us with a set of benchmarks and pretrained models. We then conduct a number of experiments using methods which are commonly used in deep learning tasks, but unexplored in their domain-specific applications for agriculture. Our experiments guide us in developing a number of approaches to improve data efficiency when training agricultural deep learning models, without large-scale modifications to existing pipelines. Our results demonstrate that even slight training modifications, such as using agricultural pretrained model weights, or adopting specific spatial augmentations into data processing pipelines, can significantly boost model performance and result in shorter convergence time, saving training resources. Furthermore, we find that even models trained on low-quality annotations can produce comparable levels of performance to their high-quality equivalents, suggesting that datasets with poor annotations can still be used for training, expanding the pool of currently available datasets. Our methods are broadly applicable throughout agricultural deep learning, and present high potential for significant data efficiency improvements.

Simulation of Automatically Annotated Visible and Multi-/Hyperspectral Images Using the Helios 3D Plant and Radiative Transfer Modeling Framework

“…Rapid measurement of traits associated with these processes has the potential to enhance our understanding of their distribution and interactions with canopy structure. In comparison to many analytical tools for plant phenotyping such as PlantCV [ 51 ] and HSI-PP [ 52 ] for plant image analysis, and AgML [ 22 ] and CropSight for data management [ 53 ], the current framework addresses the need for synthetic imagery and offers a solution for multimodal analysis tools. Compared to the LESS [ 20 ] and DART-Lux [ 27 ] models, the current framework offers integrated models for generating and modifying model geometry via Helios, making it especially suited for proximal remote sensing applications.…”

“…Although several publicly available annotated image datasets for agricultural applications are available for machine learning model training and other plant phenotyping applications, such as the Annotated Crop Image Dataset [ 21 ], Michigan State University Plant Imagery Dataset (MSU-PID) [ 13 ], AgML [ 22 ], and KOMATSUNA dataset [ 15 ], these data repositories are not sufficiently broad to capture the extensive variability that exists within agricultural machine learning tasks. The limitations of machine learning approaches become evident with small and low-variation datasets, as they can lead to severe overfitting, and the resulting models are often not readily transferable across different light conditions, species, or phenotyping platforms, revealing a lack of generalization and posing a substantial risk of extrapolation errors [ 2 , 22 ]. Past researchers have utilized data augmentation methods like random cropping, scaling, rotation, and flipping in the spatial domain [ 21 , 23 ], and introduction of random variations in mean offset and slope of the spectral reflectance [ 24 ].…”

More eyes on the prize: open-source data, software and hardware for advancing plant science through collaboration

“…The AgML project ( https://github.com/Project-AgML/AgML ) is an open-source platform designed to assist the training of machine learning models in agriculture, providing a standardised format for managing data and models ( Joshi et al , 2022 ). A common platform, specific to agriculture, with a consistent data management, training and evaluation pipeline would greatly benefit the efficiency of weed recognition and plant phenotyping DL research and development.…”