Investigation of early symptoms of tomato leaf disorder by using analysing image and deep learning models

Despite rapid population growth, agriculture feeds everyone. To feed the people, agriculture must detect plant illnesses early. Predicting crop diseases early is unfortunate. The publication educates farmers about cutting-edge plant leaf disease-reduction strategies. Since tomato is a readily accessible vegetable, machine learning and image processing with an accurate algorithm are used to identify tomato leaf illnesses. This study examines disordered tomato leaf samples. Based on early signs, farmers may quickly identify tomato leaf problem samples. Histogram Equalization improves tomato leaf samples after re sizing them to 256 × 256 pixels. K-means clustering divides data space into Voronoi cells. Contour tracing extracts leaf sample boundaries. Discrete Wavelet Transform, Principal Component Analysis, and Grey Level Co-occurrence Matrix retrieve leaf sample information.


Introduction
The domestication of primary food crops and animals today can be attributed to the advancement of agriculture several millennia ago.Food insecurity is a significant global challenge confronting humanity, with plant diseases representing a primary contributing factor.As per a scholarly estimate, plant diseases are responsible for causing a reduction in crop yield of approximately 16% on a global scale [3].According to research, the estimated global loss potential for wheat due to pests is about 50%, while for soybean, it is estimated to be between 26-29%.This information is supported by reference [3].Plant pathogens may be broken down into many distinct categories, such as fungi, fungal-like organisms, bacteria, viruses, viroid, viruslike organisms, nematodes, protozoa, algae, and parasitic plants.Applications in the fields of renewable energy power prediction [4,5], biomedicine [6,7], and computer vision [8,9] have all benefited greatly from the use of AI, ML, and CV.As a consequence of the COVID-19 epidemic, there has been an increase in the use of AI for the diagnosis of lung-related diseases [8,9,10,11] and other predictive applications [12].The implementation of comparable cutting-edge technology may alleviate plant ailments' unfavourable impacts through prompt identification and assessment during their initial phases.Currently, a significant amount of research is being conducted on the utilisation of artificial intelligence and computer vision for the automated detection and diagnosis of plant diseases.This is because manual plant disease monitoring is a laborious, timeconsuming, and arduous task.
Plants play a crucial role in human existence as they are responsible for the production of sustenance and protection against hazardous radiation.The presence of plants is essential for maintaining all terrestrial life and the safety of the ozone layer, which serves as a filter for ultraviolet radiation.The tomato plant is a nutrient-dense crop commonly grown and consumed as a vegetable [1].Globally, an estimated 160 million tonnes of tomatoes are consumed annually [2].The tomato is Despite rapid population growth, agriculture feeds everyone.To feed the people, agriculture must detect plant illnesses early.Predicting crop diseases early is unfortunate.The publication educates farmers about cutting-edge plant leaf diseasereduction strategies.Since tomato is a readily accessible vegetable, machine learning and image processing with an accurate algorithm are used to identify tomato leaf illnesses.This study examines disordered tomato leaf samples.Based  regarded as a source of income for farm households and is recognized as a significant contributor to poverty reduction [3].The cultivation and production of tomatoes notably impact the agricultural economy due to their high nutrient density, making them one of the most valuable crops globally.The tomato is considered a nutrient-dense food item, and it has been found to exhibit pharmacological properties that can safeguard against various ailments, including hypertension, hepatitis, and gingival bleeding [1].The growing utilization of tomatoes has led to a surge in demand for this commodity.As per statistical data, it has been observed that small-scale farmers are responsible for generating over 80% of the agricultural yield [2].However, they face a significant challenge in crop loss, which amounts to approximately 50% due to the impact of diseases and pests.Identifying field crop diseases and parasitic insects is crucial in determining their effects on tomato growth.Therefore, research on field crop disease diagnosis is necessary.

Proposed Methodology
This section presents a model for detecting leaf disease developed using the IP and ML approaches.The proposed model for detecting leaf diseases and utilizes a combination of computer vision and machine learning techniques, specifically DWT, PCA, GLCM, and CNN.
Tomatoes with six different diseases are used to see how well they can detect and categorize leaf diseases.To ensure that the tomato samples are the same size throughout the experiment, they are scaled to 256 256 pixels as part of the picture processing.HE and K-mean are clustering to improve the quality and divide the leaf samples.The presence of illness in a leaf may be anticipated early in the process using K-means clustering.Using contour tracing, we can identify the limits of the leaf samples.Samples' functional regions and characteristics are extracted using the DWT, PCA, and GLCM.The second step, also part of machine learning methodologies, involves classifying the features and keeping track of the model's performance using SVM, KNN, and CNN.

Development of Datasets
Tomato plants in the village database that many illnesses have hit are considered.To conduct the tests for leaf disease identification we capture photos of a tomato leaf affected by six different diseases.Figure 1 displays several representative database leaf pictures.Images of leaves clustered by using the Kmeans algorithm.K-means clustering determines how far each cluster's centre is from its neighbours, applied to each image.Figure 2 displays leaf samples clustered using the k-mean technique.
The process of contour tracing and utilized on digital leaf specimens to extract their overall shape characteristics.Upon completion of the contour extraction process, its distinct features are meticulously examined and subsequently used for pattern classification.Frequently aids the assessment of feature extraction process efficiency.Figure 3 displays the images that were generated after executing contour tracing.

Discrete Wavelet Transform
The Discrete Wavelet Transform (DWT) technique extracts significant features from improved tomato specimens.The Discrete Wavelet Transform (DWT) partitions the signal into sub-bands that consist of components with lower frequencies (LL, LH, HL) and higher frequencies (HH).According to Figure 4, the low-low part of the discrete wavelet transform (DWT) exhibits a greater degree of information availability in comparison to its higher frequency counterparts.Principal Component Analysis extracts the most valuable features from a wavelet decomposition of a grey-level cooccurrence matrix.The GLCM uses higher-order grey values determined by nearby criteria in its distribution.The GLCM method is used to extract characteristics from leaves, and this is where the various attributes come from.
These are the most common texture-based features utilized today.A sum of squares, Mean, Standard Deviation, Autocorrelation, Dissimilarity, Entropy, and Homogeneity.A feature vector composed of the parts collected using DWT, GLCM, and PCA is concatenated to recognize and categories pictures.

Machine learning Models
Samples are classified using several methods including SVM, KNN, and CNN.The Convolutional Neural Network is an example of an ANN created for data analysis.Input (IL), output (OL), and hidden (HL) layers make up CNN's complex architecture.Convolutional layers, a RELU layer that executes the activation function, pooling layers, a fully connected layer, and a normalization layer are all part of the HL.Its design is cross correlation rather than convolution, as shown by the fact that the matrix indices have mathematical importance.In Fig. 5, we see a typical -3-layer network.

Result
The suggested model will be evaluated using data collected from tomato leaf samples in a village dataset.One hundred leaf specimens in good health are used to put the proposed model through its paces.Using this model, we could correctly classify 99 samples at the 99% level.The model correctly identifies 100 out of 100 tomato samples infected with the Mosaic virus.
The model achieves perfect accuracy in the field of leaf moulds.The programme accurately predicted 99 out of 100 instances of yellow curl.Spotted spider mite and Target Spot provide identical success rates of 99% and 100%, respectively.The suggested model is evaluated by testing 685 samples from the tomato village dataset, with an impressive accuracy of 99.1%.
Training and testing the model on subsamples of the dataset ensures its accuracy.The following are the recommended system requirements for running the programme.In other words, Windows 10 operating system, GPU-NVIDIA core, and Python programming language.OpenCV, Keras, tensor flow, NumPy, Matplot, MatPlot, and the image data generator,Tomato (Six Disorder) samples from the Plant Village collection.Precision, Recall, and F1 score measure the effectiveness of the suggested model.The outcomes of the proposed model are contrasted with those of pre-existing models.The study indicates that the proposed model, which combines DWT, PCA, GLCM, and CNN, exhibits a higher level of accuracy at 99.09% in comparison to other pre-existing models.

Conclusion:
In the suggested model, computer vision methods are used for preprocessing, including converting from RGB to greyscale, Histogram Equalisation (HE), K-means clustering, and contour tracing.Feature extraction methods that are renowned for their capacity to extract useful features are applied to the leaf samples.These methods include the Discrete Wavelet Transform, Principal Component Analysis, and the Generalised Linear Model.To distinguish between healthy and diseased leaves, researchers use machine learning methods including Support Vector Machines (SVM), K-Nearest Neighbours (K-NN), and Convolutional Neural Networks (CNN).The proposed model has been shown to be suitable for the CNN machine learning classification methodology, with its recommended degree of accuracy when compared to other modern methods.Using fusion approaches to extract significant characteristics and testing the model on more leaf sample datasets might improve future research.

Figure 1 :
Figure 1: tomato leaf with different diseases

Figure 5 :
Figure 5: ANN Model -Input, hidden and output layer

Fig. 6
Fig.6shows the CNN confusion matrix for output and target classes.Fig.7shows CNN-classified leaf feature training samples and their accuracy and mistakes.

Figure 8 :
Figure 8: Comparison of proposed Deep learning clustering+GLMC+SVM YC K-Mean clustering+GLMC+SVM LM K-Mean clustering+GLMC+SVM MV K-Mean clustering+GLMC+SVM Health EAI Endorsed Transactions on Internet of Things | Volume 10 | 2024 | Investigation of early symptoms of tomato leaf disorder by using analysing image and deep learning models on early signs, farmers may quickly identify tomato leaf problem samples.Histogram Equalization improves tomato leaf samples after re sizing them to 256 × 256 pixels.K-means clustering divides data space into Voronoi cells.Contour tracing extracts leaf sample boundaries.Discrete Wavelet Transform, Principal Component Analysis, and Grey Level Co-occurrence Matrix retrieve leaf sample information.

Table 1 :
Table 1 compares SVM, KNN, and CNN based on their precession, recall, and F1 scores.Comparison of different Deep learning model combinations