Embedded Files
2020 - SICECA, Portable Software Tool to Identify and Classify Sugar Cane Diseases
Currently, the process that is carried out for the diagnosis of diseases in sugarcane starts with the cultivator, who must perform a complex process for its diagnosis. In response to this problem, SICECA is a system to identify and classify diseases such as Roya café, Mancha anillo, Mancha púrpura y Muermo rojo using the Faster R-CNN architecture, which eases the diagnosis of a sick sugarcane plant through images of its leaves.
The methodological development is based on ANN, methods and techniques used to solve this kind of problem. Lately, a procedure of annotation and depuration was performed to a private dataset of images with diseased leaves on uncontrolled environments, such that the images with appropriate conditions were selected.
Actually, SICECA is a registered software legally recognized by the Interior Ministry of Colombia. This document can be downloaded here.
The development of SICECA had two main stages, which are shown in the figure at right.
First stage:
The dataset used was prepared and checked to debug any error present.
Also, methods of data augmentation were used to increase the amount of data.
Furthermore, the Faster R-CNN architecture was adjusted in order to be trained accordingly with the dataset available.
This process was iteratively done until the Faster R-CNN got good enough results.
Second stage:
Once the Faster R-CNN was trained, SICECA GUI is used from user point of view in order to detect and classify sugar cane diseases in the field.
In the field, users can introduce pictures and know what kind of disease (Roya café, Mancha anillo, Mancha púrpura or Muermo rojo) is present in the sugar cane plant.
In addition, users can introduce many images and process them in batch.
The second stage was implemented in the SICECA GUI, specifically using two modules the processing data and validation modules.
The processing module is shown in the figure at left, it allows users to perform the following functionalities:
Loading the image to process.
Loading the set of images to process.
Selecting a folder where the images are stored.
Loading the Faster R-CNN architecture trained.
Processing the images to detect and classify sugar cane diseases.
Observing the result of the images processed labeling the region on the image where the disease was detected.
Selecting the folder where the results will be exported.
In addition, the validation module of SICECA allows users to show many statistics about the detection and classification procedure performed.
Here, there are two cases. First, when users have the ground truth. And second, when users do not have the ground truth.
In the second case, which is usual when users are employing SICECA in the field. In this case, SICECA can show the following statistics:
Precision.
Accuracy.
In the first case, which is usual when users are validating other neural network architecture. In this case, SICECA can extract the following statistics:
Confusion matrices.
Mean IoU by class.
Precision
Accuracy
Recall
F1 score
Also, in the validation module results can be exported.
This software was developed for industrial service and academic use. If you like to perform a field test, please contact me:
Prof. Bladimir Bacca Cortes Ph.D.
Address: Cra. 100, Street 13, Universidad del Valle, Melendez, Building 354, Office 2006.
Tel: +5723212100 Ext. 7656
Page updated
Report abuse