This project presents a plant image classification scheme that uses a combination of Unet-based image segmentation and a convolutional neural network (CNN) architecture for the actual classification. The first step of the proposed approach is to segment the plant leaves from the background using a modified Unet architecture,which is a popular deep-learning model for image segmentation. The segmented leaves are then preprocessed and fed into a CNN architecture for the actual classification. The CNN architecture consists of multiple convolutional layers, followed by pooling and fully connected layers, which enable the model to learn the complex features necessary for accurate classification. To evaluate the proposed approach, experiments were conducted on a publicly available Plant village dataset. The results show that the proposed approach achieves high accuracy in classifying different plant species. The dataset used for the project can be downloaded from here
The percentage of crop yield lost to plant diseases each year varies depending on the crop, the region, and the specific plant disease. According to the Food and Agriculture Or- ganization (FAO) of the United Nations, plant diseases and pests are responsible for the loss of up to 40% of global food crops each year. In some cases, the percentage of crop yield lost to plant diseases can be much higher. For example, some estimates suggest that up to 80% of banana crops worldwide are at risk from the fungal disease known as Panama disease, also called Fusarium wilt, which can cause complete crop loss. It’s worth noting that losses due to plant diseases can also have significant economic im- pacts beyond just the loss of crops, including increased costs for control measures and reduced market access for farmers. Early detection of plant diseases remains challenging due to lacking lab infrastructure and expertise.
This project uses the Plant village dataset found in (Gomaa, 2023).The Plant Village dataset is a collection of over 54,000 high-quality images of 14 different crop species, including tomato, potato, apple, and grape. Each image is associated with one of 38 different classes, representing various plant diseases or healthy conditions.
The Python OpenCV library provides several functions for image thresholding, a process that converts grayscale or color images into binary images. In this process, each pixel in the image is compared to a threshold value and is assigned a binary value (0 or 255) based on whether it is above or below the threshold value. The OpenCV library is used in this project to convert the RGB images into segmented images with a black background, but it failed to give decent results for all the images, as shown in Fig.3. A qualitative study of the results of this computer vision scheme showed an efficient performance of this strategy on the tomato healthy leaves images. Consequently, the obtained tomato healthy binary images are used as the ground truth for training the UNET (Ronneberger et al., 2015) image segmentation strategy.
The U-net is an advanced deep learning architecture designed for image segmentation, focusing on biomedical image analysis (Ronneberger et al., 2015). Its name is derived from its U-shaped network architecture, distinguishing it from conventional CNN models. In contrast to standard CNN models, U-net employs convolutional layers to upsample or combine feature maps into a complete image. This project introduces the conventional U-Net architecture utilized in (Vitali, 2020) using the tomato healthy as a baseline training subset. In addition, to reduce the number of channels of each image, OpenCV is newly used to obtain grayscale images for faster training performance. The architecture utilized in this work can be seen in the image below.
