Object detection (OD) technology, which identifies and classifies objects in images and videos, has been widely adopted across various fields. However, implementing OD faces challenges, including image preprocessing, labeling, model development, and deployment. To streamline these processes, we developed a Python-based software Ladder (Labeling and Detection Deployment for Entity Recognition). Ladder features a user-friendly graphic interface (GUI) that facilitates efficient labeling of training datasets, detection of new images, and model training. The software utilizes an interactive recurrent framework that begins with predictions from a pre-trained model for initial image labeling. Users can then add human labels, and these newly labeled images can be incorporated into the training data to retrain the model. In this study, we demonstrate an efficient development of a broken rice detection model using Ladder. The model employed a three-stage training process and demonstrated strong predictive performance (R2 = 0.99), with a mean absolute error (MAE) of 6.08 (95% CI: 5.18–6.97) and a root mean square error (RMSE) of 6.68 (95% CI: 5.93–7.46). Rice is one of the world’s most essential crops, with the rate of broken rice significantly affecting its price in the market and potential uses. This necessitates an efficient method for assessing the ratio of broken rice for breeding, production, and trading.
Loading....