Adapted Binary Thresholding for Automated Seedling Estimation

บทคัดย่อ

Accurate seedling counting in trays is essential for optimizing planting area preparation in agricultural production. This study presents an Adapted Binary Thresholding (ABT) method for automated seedling counting, leveraging image processing techniques. A case study was conducted on Sida tomato seedlings, where images were captured using a smartphone camera under natural lighting conditions. The dataset consisted of eight images from eight 105-hole seedling trays, with the first four tray images allocated for training and the left four for testing. The ABT method uses an RGB-based segmentation to distinguish seedlings from the background. Following segmentation, morphological operations, and connected component analysis were applied to refine the seedling count. The algorithm was implemented using Octave, an open-source software with a nice GUI suited to test an algorithm. Parameter optimization was performed during the training phase, followed by evaluation on the test set. Performance was assessed using hit rate, merge rate, split rate, false alarm rate, accuracy, precision, recall, and F-score. The proposed method achieved a high hit rate of 94.55% on the training set and 94.47% on the test set, with minimal split and false alarm rates. The precision, recall, and F score of the training set were 1, 0.9455, 0.9720, and those of the test set were 1, 0.9447, 0.9716, respectively. These results demonstrate the effectiveness of ABT as a lightweight, high-accuracy image processing technique suitable for real-time applications in resource-constrained environments, such as smartphone-based agricultural tools.