DETECTION OF LEAF SPOT DISEASE IN OIL PALM SEEDLINGS USING CONVOLUTIONAL NEURAL NETWORK METHOD

: This research aims to develop a method for detecting leaf spot disease in oil palm seedlings using Convolutional Neural Network (CNN). Leaf spot disease in oil palm seedlings can hinder growth and production. CNN has proven effective in image processing and classification, particularly in plant disease detection. In this study, we utilized a dataset of images containing oil palm seedling leaves infected with leaf spot disease and healthy leaves. We performed data processing, built a CNN model, and conducted hyperparameter tuning. The test results demonstrate that the developed CNN model achieves high accuracy in recognizing and distinguishing between oil palm seedling leaves infected with leaf spot disease and healthy ones. This research contributes to the development of plant disease detection technology that can support economic growth in the oil palm plantation sector.


INTRODUCTION
Indonesia is a country rich in plantations, plantations play an important role in Indonesia's economic growth.Currently, palm oil has become one of the main commodities in Indonesian agriculture that plays an important role in generating foreign exchange revenue.The palm oil industry sector has a growing market share, reaching an increase of 9.23% in 2017.In addition, the industry is also the largest contributor to foreign exchange earnings from the non-oil and gas sector, reaching around 34.33% in the same year.According to data from the Directorate General of Plantations of the Ministry of Agriculture, the area of oil palm plantations in Indonesia has expanded to reach 8.9 million hectares [1].
Oil palm plantation development relies heavily on nursery activities [2].Seedling growth is a key factor that can affect the success of oil palm production in the field [2].In nursery activities, it is important to pay attention to the presence of diseases that can attack seedlings [3].Therefore, it is important to study the incidence of disease in plants in order to effectively control the disease, ensure oil palm seed production runs smoothly, free from disease, and optimal growth.One of the most common diseases on oil palm seedlings is leaf spot disease [4] [12].The objective of the study was to recognise whether the palm leaves were affected by disease or not through a programme.An accurate approach is required to achieve a high level of accuracy.The researcher used the Artificial Neural Network (ANN) classification method along with the Local Binary Pattern (LBP) extraction method.The first step taken on the image was to convert it to grayscale, and then proceed with the extraction using the LBP method.The results were analysed through a JST with the application of 17 training functions resulting in 5 neurons.The results show an average accuracy of about 81%, a precision accuracy of about 95%, and a recovery rate of about 94%.On using 10 neurons, the accuracy increased to an average of 95%, with a precision of about 97%, and a recovery rate of about 96%.When using 20 neurons, the average accuracy reached 97%, with a precision and recovery rate of about 97% and 96% respectively [12].
Referring to previous research [12], There are similarities in the focus of the discussion, namely disease detection in oil palm leaves.However, the significant difference lies in the method implemented and the emphasis on one type of palm leaf disease.The method applied in this research is Convolutional Neural Network (CNN).The dataset used consists of images of oil palm seedling leaves that have spot disease and healthy oil palm seedling leaves.This dataset is obtained by direct collection.The final result expected from this research is to achieve the highest accuracy, precision, and recall values.

METHOD
The method used is the Convolutional Neural Network (CNN) method.The flow diagram of the research methodology can be seen in Image 1.

Image 1. Flow of Research Methodology
The first stage of conducting a literature study, in general, a literature study is a method for overcoming problems by investigating references to previously existing writings.The literature study in this research involves searching for journals that are relevant to the research topic to be carried out.The next stage is data collection, Data collection is a phase in the research process where researchers use certain scientific methods and techniques to collect data in a structured manner for analysis purposes.In this study, the dataset was collected directly from PPKS Marihat, Sei Kumango Village, Tambusai, Rokan Hulu Regency, Riau Province.At the data processing stage, the dataset will be divided into two, namely training data and validation data.After the data is divided into two, normalise and resize the dataset.Continue with the construction of the CNN model, the structure used involves convolution layers, maximum fusion, dropout, and output layers, with the aim of producing a model that is effective in recognising features on image datasets.This research will end by presenting the results and conclusions obtained.

Dataset
The dataset used in this study consists of leaves of oil palm seedlings affected by spot disease and leaves of oil palm seedlings that are normal or not affected by disease.Dataset collection was carried out directly from PPKS Marihat, Sei Kumango Village, Tambusai, Rokan Hulu Regency, Riau Province, for palm leaves affected by spot disease using a Canon camera and for normal palm leaves using an Oppo Smartphone camera.As can be seen in Image 2, there is a sample dataset of normal palm seedling leaves and a dataset of palm seedling leaves affected by spot disease.The total dataset consists of 600 palm seedling leaves, with 300 seedlings infected with spot disease and 300 healthy palm seedling leaves.The dataset will be divided into two parts, 70% for training data and 30% for validation data.

Preprocessing
Preprocessing is a stage or step in data analysis that aims to improve the data and prepare it to be processed by an algorithm or model.By preprocessing, data can be transformed into a form that is more suitable for further analysis, such as removing empty data, changing the data format, or normalising the data scale, so as to produce cleaner data that is ready to be used in the data labelling and modelling process [13].After the data is divided, the next step is to preprocess the dataset with normalisation using the Rescale 1/255 function.This normalisation rescales the image so that the pixel values are in the range of 0 to 1.In addition, the size of the target image to be used in the training process is 600 x 400 pixels.

Convolutional Neural Network
Convolutional Neural Network (CNN) is a type of Deep Learning algorithm that was created specifically to recognise and process images or image input.This algorithm has the capability to extract important information or learnable weights from various aspects or objects in the image, and is also able to separate and identify each object separately [14].The feature extraction process in the CNN model is executed automatically through a convolution process, which consists of a convolution layer and a pooling layer [15].

CNN Model Development
The next stage is the construction of the CNN Model, the Convolutional Neural Network structure used in this study can be seen in Table 1.Pooling layers are used to reduce the dimensions of feature maps, especially for width and height, while maintaining depth.Max pooling is a type of pooling that produces maximum values within the region of the image covered by the filter.Max pooling is more suitable for extracting dominant features, and is therefore considered better in some cases.In the process of building the model, there are several stages consisting of convolution blocks and max pool layer.To avoid overfitting which can lead to poor predictions on the dataset, the Dropout layer technique is used.Dropout is a technique in CNN that helps overcome the dependency problem between neurons [16].Next, the 'Flatten' layer is processed, which aims to convert the input into a single vector without parameters, so that it will not affect the batch size in the dataset training process.

Hyperparameter Tuning
Hyperparameter tuning is a crucial aspect in the world of machine learning to achieve maximum accuracy.The process of determining the optimal configuration in hyperparameter tuning involves repeated iterations with various combinations of different configurations.[17].The aim is to select one configuration that can provide the highest accuracy.By repeating and varying the configuration, it is expected that the model can be optimally adjusted so that it can provide accurate predictions on new data.Through the hyperparameter tuning process, the model can be optimised and able to produce better performance than using the default configuration [18].

RESULTS AND DISCUSSION
In this study, the Convolutional Neural Network (CNN) method is used to detect spot disease in oil palm seedlings.The dataset used consists of 600 images of oil palm seedling leaves, with 420 images used as training data and 180 images as validation data.The next process is preprocessing the training data and validation data, which involves Resize, Data normalisation, and Encoder label transformation.After preprocessing is complete, the next step is to build a CNN model for spot disease detection on the leaves of oil palm seedlings using the Python programming language.The CNN model architecture is built using a sequential model and Adam as an optimiser.After the CNN model is built, testing is carried out by applying the model that has been made.Furthermore, the calculation of Accuracy, Recall, Precision, and F1-Score values is carried out.The test results show the accuracy value of the CNN model that has been built with 30 epochs of 0.99, recall of 1.00, precision of 0.99, and f1-score of 0.99.The accuracy value that is close to 1 indicates that the performance of this model is very good in image matching or spot disease detection on oil palm seedlings.The higher the accuracy value, the more accurate the model is in predicting and classifying images between the leaves of oil palm seedlings affected by spotting and normal leaves.The high recall, precision, and f1-score values also indicate that the model can properly identify and distinguish between the disease class and the normal class.This shows that the CNN model built is very effective and reliable in detecting spot disease in oil palm seedlings.
(a) Normal leaves of oil palm seedlings (b) Spotted leaves of oil palm seedlings Image 3. Accuracy Plot The accuracy result graph of training and validation shows the x-axis represents the number of epochs in training the model from 0 to 30, while the y-axis represents the accuracy level from 0.75 to 1.00.The graph in Image 3 shows excellent performance.It can be seen that the training accuracy increases rapidly as the number of epochs increases, while the validation accuracy also increases.Image 4. Loss Plot The graph of loss results from training and validation displays the x-axis representing the number of epochs in training the model from 0 to 30, and the y-axis representing the loss rate from 0.5 to 0.0.The graph in Image 4 shows excellent results.It can be seen that the loss in training decreases as the number of epochs increases, and the loss in validation also decreases consistently.This indicates that the model is working well and getting closer to the expected results as the training process progresses.Image 5. Classification Report