Cr'Hope (Hope for Crop)

I am very happy to have been able to participate in the CR’HOPE (hope for crop) project which won the Grand Prix du Jury at the Hackatech organized by Inria Sophia Antipolis, with a strong team composed of Paul GANELON, Jérôme JACQUES, Rémi Jolin and Chems-Eddine Ouaari!

This project, formerly called Insectae, is led by Jérôme JACQUES and proposes a solution of connected and intelligent insect trap alerting farmers of the presence of pests that cause on average more than 10% of production losses. This solution allows a punctual and parsimonious treatment of plots, thus reducing the use of phytosanitary products. The economic, ecological and societal impact of this project is obvious, and the PoC that we developed during these 48 hours has demonstrated its feasibility.

Algorithms and technologies used in this PoC:

• Visual detection and extraction of insects on the attractive trap using classical computer vision methods (thresholding, contours) in Python and OpenCV

• Classification of the detected insects using a neural network (CNN ResNet) fine-tuned on the IP102 dataset containing 102 insect species. Our network reaches an accuracy of 73% on these 102 classes and 98% on our main use case which is the classification of leafhoppers. Use of Python, PyTorch, TensorRT.

• Estimation of pest density in the entire agricultural plot from a limited number of acquisition points using the Shepard interpolation method (Python and Numpy).

• Projection of pest proliferation in the plot as a function of time by implementing a reaction-diffusion simulation (Fischer equation). Python, SciPy, Numpy.

The whole was implemented on a Nvidia Jetson Nano card (max. power consumption 15W) powered by a solar panel.

Thanks to Jérôme JACQUES for accepting me in your team and good luck for the rest of your project !

Thanks also to Pierre Alliez for having pushed me to register to this event!

See this linkedin post for more photos.