Abstract:
Common bean is one of the most important crops produce by farmers in Ethiopia for export and
local consumption purpose. However, the quality and productivity of this crop is highly affected
by different factors including leaf diseases which affect the plant during its growth. Currently
common bean diseases detection is doing manually through a visual inspection of an expert.
However, visual based crop diseases detection is expensive, time consuming and inaccurate.
Therefore, a deep learning approach is proposed to detect the disease. This research focus on the
three leaf diseases of the common bean namely, common bean brown spot, common bean rust
and common bean beetle disease. For this study, with the help of farm area expert 1766 images
of common beans were captured using smart phone camera from common bean farmlands. On
leaf images, we do color based segmentations. The proposed model has four convolutional layers
and each layers followed by max pooling layers, activation layers and batch normalization
layers. We are used softmax classifier to classify the leaf diseases. The experimental results show
that the proposed model performs better than pre-trained models. It scored 98.9%, 97% and 96%,
training, validation and testing accuracy respectively. Moreover, the three Deep Learning models
namely, AlexNet, GoogleNet as InceptionV3 and VGG16 were trained with the same simulation
environment to made comparison with the proposed model. The experimental results show that the
proposed model achieves better accuracy than the pre-trained models. Thus, future works is
directed on automatically estimating the severity of the detected leaf disease and increasing the
number of leaf diseases for provides a feasible solution for detection of common bean leaf
diseases
