The proposed pipeline in our paper is based on a deep convolutional network to classify Arabidopsis thaliana plant parts into 4 classes and use patch based classification approach (through sliding window method) to detect and count fruits of this plant.
This repository contains the following folders:
-
Annotation-Toolbox : A GUI for manually annotating plants
-
prepare data to train model : To prepare 3 datasets (training, validation and testing) used for developing the classification model.
-
trained models: Two convolutional neural networks were trained and the trained weights can be used in the Caffe platform for doing classification.
-
silique counting : reconstructs the plant image to detect and count the silique numbers on the whole plant image.
-
To do annotation refer to Annotation-Toolbox.
-
To train the model from scratch refer to prepare data to train model.
-
To use the trained models (no need for learning process) for 4-class classification, refer to trained models.
-
To count the number of siliques, and extract several quantitative phenotype information refer to silique counting.
Requirements
-
MATLAB 2017 v9.3 or above, CAFFE 1.0.0-rc3 or above.
-
Other requirements: CUDA version 8.0, CuDNN v5.1, BLAS: atlas, DIGITS version5.1-dev.