KGCV_Strawberry

A knowledge-guided computer vision framework for strawberry fruit detection and growth modeling.

For detailed methodology please refer to "Qi Yang, Licheng Liu, Junxiong Zhou, Mary Rogers, Zhenong Jin, 2024. Predicting the growth trajectory and yield of greenhouse strawberries based on knowledge-guided computer vision, Computers and Electronics in Agriculture, 220, 108911. https://doi.org/10.1016/j.compag.2024.108911"

Dataset and pre-trained models

The complete dataset and the pre-trained models can be downloaded from https://doi.org/10.5281/zenodo.10957909

Training

Train detection network

"KGCV_Strawberry_Train_FasterRCNN.py" is used to train a faster-RCNN for strawberry bounding box and main phenological stage detection

Train trait estimation network

"KGCV_Strawberry_Train_CNN.py" is for traing a CNN to estimate the fruit size and decimal phenological stage.

Train parameter network

We employ the S-shape function to represent the fruit growth progress, with its curve parameters serving as the fruit growth parameters. $$f(x) = \frac{y_0 - s}{1 + e^{-r(x-t_0)}} + s$$ where $y_0$, $s$, and $r$ are three curve parameters that control the trajectory of fruit size. "KGCV_Strawberry_Train_MLP-1_parameterNet.py" builds the mapping from a sparse 1×70 diameter/length sequence to these three curve parameters. To establish this mapping, this script uses the "loadSyntheticData" function to generate the synthetic data

# load dataset    
syn = loadSyntheticData(mode=mode,ensembleN=1000)

Then this generated synthetic dataset was used to train the parameter network.

Train fruit growth network

"KGCV_Strawberry_Train_MLP-1_growthNet.py" trains an MLP to map fruit size at $t$ to $t+\Delta t$ $$s_{t+\Delta t} = g(s_t, \Delta GDD, Para)$$

Demo

An integrated demo was provided in "KGCV_Strawberry_demo.py"