Detecting brown spot in rice leaf is an urgent complication in the agricultural field as Brown Spot disease lessen the rice yield remarkably. Several segmentation techniques have been applied to identify and extract the infected portion of the rice-leaf and machine learning algorithms such as decision trees, support vector machines are applied to detect this infection. In particular, a combination of Convolution Neural Networks with these algorithms has also tried to resolve this problem. Although this attempt has achieved success in providing accuracy (96.8%), these kinds of approaches raise issues regarding the size and interpretability of feature space and interpretability of the decision model. Indeed, Deep learning networks automatically create a feature space that usually contains a massive number of features (numerous of them are not necessarily appropriate). This vast number of features extends the non-interpretability of the machine learning model. Furthermore, training the model with these many features is computationally expensive. To resolve these issues, we propose a method to extract a few interpretable features from rice-leaf images and construct a low-dimensional feature space; however, interpretation shows that they deserve significant credit for the decent accuracy of our classification model.
Interpretable Learning Model for Lower Dimensional Feature Space: A Case study with Brown Spot Detection in Rice Leaf
1. Interpretable Learning Model for Lower Dimensional Feature Space: A
Case study with Brown Spot Detection in Rice Leaf
Presented by: Md. Amit Khan
Authors
Md. Amit Khan1 Nafiz Sadman1 Kishor Datta Gupta2 Jesan Ahammed Ovi3
1. Silicon Orchard Ltd, Bangladesh
2. University of Memphis, TN, USA
3. East West University, Bangladesh
2. Brown Spot Disease
● Caused By Bipolaris Oryzae and Cochliobolus Miyabeanus
● Major Symptom is cylindrical, circular and oval shaped brown dots on leaves
● Is cureable if detected early
3. Data Set
● Downloaded from Kaggle
● Four classes (Brown Spot, Healthy, Hispa, Leaf Blast)
● Each Class has 523 images
4. Previous Attempts
● Decision Tree with 10-fold cross validation
● SVM classifier to detect three kinds of rice diseases
● Combined approach of image processing and soft computing
● Rule Mining Approach
● Combination of a CNN model with a SVM classifier
6. 1. Extract Aggregated Features From Images
● Aggregated Laplacian Coefficient
● Estimated Brown Spot Area
● Sum of contours area
7. ● Laplacian Based features to identify the brown
Spot’s color and shape properties
● Measured for red, green and blue channels
Individually
● Aggregated laplacian coefficient by summing
each laplacian
8. Estimated Brown Spot Area
● Images are converted into binary images
● All contours are obtained
● Areas of the contours are calculated
● Second largest area of the contours are
considered to be the brown spot area
9. 2. Train The Classifier
● A random forest classifier was chosen
● Best set of parameters were chosen by grid search
● Data set was divided into two sets (training and testing)
10. Experimental Results and Analysis
Model TP FP TN FN
Logistic
Regression
89 25 9 87
Decision Tree 87 27 8 88
Random Forest 91 23 5 91
11. Experimental Results and Analysis
Comparative Performance
Model Precision Recall F-1 Score
Logistic Regression 0.84 0.84 0.84
Decision Tree 0.84 0.84 0.83
Random Forest 0.87 0.87 0.87
12. 3. Interpretation of Classifier Decision
Two types of explainer were used to provide interpretation of the model:
1. Overall Explainer
2. Local Explainer
13. Overall Explainer
Feature importance
Variable Importance
Laplacian Blue 0.380751
Laplacian Green 0.258484
Estimated Brown Spot Area 0.190786
Sum of contours Area 0.094294
Laplacian Red 0.075685
14. Local Explainer
3 Variables with the most contributions for increasing the probability of brown spot
Variable Impact
Laplacian Blue +32%
Estimated Brown Spot Area +16%
Sum of Contours Area +6%
15. Local Explainer
3 Variables with the most contributions for decreasing the probability of brown spot
Variable Impact
Estimated Brown Spot Area -14%
Laplacian Green -13%
Laplacian Red -5%
16. Future Work
● Extended method to detect Hispa and Leaf blast
● More Specific feature set
● Apply to other crops