Crop-Type Recognition in Street-level Images with Convolutional Neural Networks

• DOI: 10.20944/preprints202307.0724.v1
• Publication Date: 2023-07-13T00:08:28Z
• Authors: Fernando Orduna-c...; Marcial Sandoval-...; Ian MCCALLUM; Linda SEE; Steffen Fritz; Santosh Karanam; Tobias Sturn; Valeria Javalera-...; F. Fernando Gonzá...

The creation of crop-type maps from satellite data has proven challenging, often impeded by a lack accurate in-situ data. This paper aims to demonstrate method for (ie. Maize, Wheat and Other) recognition based on Convolutional Neural Networks using bottom-up approach. We trained the model with highly dataset crowdsourced labelled street-level imagery. Classification results achieved an AUC 0.87 wheat, 0.85 maize 0.73 other. Given that wheat are two most common food crops globally, combined ever-increasing amount available imagery, this approach could help address need improved monitoring globally. Challenges remain in addressing noisy aspect imagery buildings, hedgerows, automobiles, etc.), where variety different objects tend restrict view confound algorithms