Product and seed counting using image processing techniques

Seed is the most basic and most important link of the food chain. A decrease or deficiency in the amount of seeds on Earth will cause a very difficult situation for life. In the world, plant production, that is agriculture, started with the discovery of seeds. In the use of seeds in agriculture, the amount of seed varies according to the characteristics of the seed. In addition, it is very important to quickly calculate the yield of the agricultural products. Just like agriculture on soil, agriculture in water has started to be seen recently. Planting enough seeds in the area to be planted directly affects the yield. More or less seeding in the agricultural field reduces the yield. Today, the amount of seed to be planted in the agricultural field is made according to the weight of the seed to be used. It is an incorrect practice to determine the seed per square meter with this method. Because the density and volume of each seed differs. The correct method is to determine the amount of seed to come per unit area according to each product at the beginning. Thus, the total amount of seeds required for the planted area can be calculated optimally. In this study, the process of counting the seeds to be planted and the yield of the obtained product was carried out quickly with the computer aided image processing method. For this, images were taken in a closed environment on a fixed platform so that seeds and crops are not affected by the difference between day and night and different light intensity. In addition, it has been tried to reduce the reflection that may occur by using different ground materials. After applying thresholding, median filtering, morphological dilation and morphological erosion as additional components as image processing method, blob detection algorithm is applied. First of all, after applying the additional components one by one, the results were obtained by applying the blob algorithm. However, in the results obtained, 98.13% success was achieved by applying dilation or erosion only to chickpea, lentil and rice products, but the same success could not be obtained by only applying thresholding or median filtering. There have been erroneous counts in blob detection using only thresholding or median filtering. The R² value is 0.92 on a smooth surface, and 0.22 on a matte surface, when counting only by applying thresholding. Counting by applying median filtering, the R² value is 0.95 on a smooth surface and 0.28 on a matte surface. When these values are taken into account, it can be seen that only thresholding or median filtering can not be achieved on the smooth surface, and the error rate is high since it shows a low correlation on the matte surface. Therefore, it has been concluded that the most successful results will be obtained with the hybrid use of additional components. In the experiments, with the hybrid use of additional components, 98.58% success was achieved in product and seed counting.