In Indonesia, corn is the second most important food crop commodity after rice. Corn cultivation can increase corn production in Indonesia. In the case of corn cultivation, problems often occur. One of the problems in corn cultivation is the appearance of leaf spot disease. The results of the study showed that the yield of corn crop decreased from 5-50% due to leaf spot disease. If leaf spot disease attacks the corn plants before female flowers appear, then the yield reduction can reach 50%. Monitoring the health of maize leaves on large areas takes a lot of time and becomes a difficult task. Currently, drone technology is an alternative to monitor the health of corn plants. Drones with cameras can be used for taking leaf images. The resulting leaf image is then analyzed further using a computer to detect its health.

 

Image segmentation is a technology that can be utilized to detect healthy corn leaves and those infected with leaf spot disease. Early detection of leaf spot disease can be used to treat this disease early so that it can increase corn production. However, segmentation of leaf spot disease in corn plants is not an easy task.

There are several image segmentation techniques, namely thresholding-based, edge-based, clustering, and neural network-based. Among these techniques, the easiest and most widely used technique is clustering. One of the clustering methods is K-means. It is easy to implement and has a relatively small computing time.  K-means has strong exploitation because it is able to focus searches around areas that are considered optimum. However, K-means clustering is quickly converging to the local optimum or slowly converging to the global optimum, and very sensitive to the initial cluster center. K-means is easily trapped around the local optimum value, as a result of improper segmentation.  Contrary to K-means, Particle Swarm Optimization (PSO) is a global search algorithm that has strong exploration, because it is able to search all solution areas to get optimum values. The conventional PSO is difficult to find inertial weight constants and appropriate social parameters.  To overcome this problem, Dynamic Particle Swarm Optimization (DPSO) is an improved PSO. However, DPSO still has the same problem with the conventional PSO, it is slow to converge to local optimum. Dynamic Particle Swarm Optimization and K-means (DPSOK) has been proposed to balance exploration and exploitation so as to obtain a better algorithm.  For this reason, this paper proposes an automatic segmentation corn leaf spot disease by using DPSOK.

 

This research has several steps which are literature study, data collection, program implementation. program testing and evaluation and conclusions making. The automatic segmentation corn leaf spot disease proposed has several parts which are input leaf image, a transformation of the RGB color space image into the HSV color space, dimension, transformation of H, S and V matrices, and an automatic segmentation corn leaf spot disease using DPSOK.

Figure 1 shows an input image, a leaf spot disease segmentation by K-means, a leaf spot disease segmentation by DPSOK and a manual segmentation of leaf spot disease. It is shown that DPSOK compared to K-means is more accurate. K-means has much miss detection of a leaf spot disease area.

Table 1 shows the numerical evaluation of the K-means and DPSOK. DPSOK has a smaller average fitness value compared to  K-means.  Table 1 also shows that the fitness value standard deviation of DPSOK is smaller than the fitness value standard deviation of K-means. It means that the segmentation result of  DPSOK is better if it is compared with K-means. However, DPSOK is more take time.