Predicting Tulip Leaf Diseases: A Integrated CNN and Random Forest Approach

In this paper, the methods of Random Forest (RF) & Convolutional Neural Network (CNN) algorithms are combined to present a novel approach for classifying bacterial infections in plant species. The proposed model extracts characteristics from the input photos using three layers of convolution, three layers that max-pool features, and a layer that is completely connected before categorizing the images using RF into various bacterial disease classifications. Recall, F1-Score, support, and accuracy measures were used to evaluate the model’s performance using a dataset of plant image data that included seven different types of bacterial illnesses. The experimental results demonstrate that the proposed CNN-RF methodology provides remarkable precision in categorizing the images when compared to other state-of-the-art techniques. The model specifically achieves an average weighted recall of 81.54%, an F1-Score of 81.54%, and an accuracy of 81.54%. The recommended approach is effective at identifying various bacterial illnesses in plants because the model outperforms rival models when it comes to of macro & micro-average F1-Scores. The recommended method for diagnosing bacterial diseases in plants makes use of machine learning as well as deep learning techniques. Because bacterial illnesses in plants can have a substantial influence on agricultural production, a model’s high precision, and efficacy can help identify and stop the spread of these infections.

I. Introduction

The gorgeous and extensively cultivated tulip is cultivated across the world. Tulips are susceptible to several illnesses that can stunt their growth and reduce their output, much like other plants. To reduce financial losses, these diseases must be promptly detected and managed. Machine learning methods have recently shown potential in the diagnosis of plant diseases. In this study, we present a combination method for predicting tulip leaf illnesses using Convolutional Neural Networks, or CNN, and random forest (RF) algorithms. Convolutional neural networks based on artificial intelligence (CNN), a type of deep learning model, have excelled in applications requiring picture identification. Four convolutional layers plus four max-pooling layers make up the CNN model. The max-pooling layers lessen the depth of the resulting feature maps while the convolutional layers extract features from the input image. We use the corrected linear unit (ReLU), which is the function of activation in each convolutional layer to add nonlinearity to the model. Three fully connected layers receive the output features following the convolutional plus max-pooling layers. We adopt a regularisation strategy in the final fully linked layer to prevent the model from overfitting. By including a penalty component in the loss function, regularisation aids in the reduction of the model’s variance. We employ the weight decay method of L2 regularization, which imposes a penalty equal to the square value of the weighted values. The softmax layer, the last stage of the CNN model, outputs the expected probability of the various tulip leaf illnesses. The Random Forest (RF) classifier uses an ensemble of decision trees to increase classification performance after receiving the probability estimates from the CNN model. The RF method creates several decision trees using a selected number of input attributes. The results of all of the choice trees are averaged to produce the final result. Because of their great precision and noise resistance, RF algorithms are well recognized. In this study, we apply the suggested method and assess its effectiveness using a dataset of pictures of tulip leaves with various illnesses. Our tests’ outcomes demonstrate that the suggested strategy exceeds current state-of-the-art methods concerning accuracy and effectiveness.