<![CDATA[Fruit image classification plays a crucial role in smart agriculture, particularly in automating sorting and quality control processes. This study proposes a fruit classification system by integrating HSV color space conversion, adaptive thresholding, morphological segmentation, and the Extreme Learning Machine (ELM) algorithm. The dataset consists of three fruit classes—apple, pineapple, and watermelon—with a total of 480 images, divided into 360 training samples and 120 testing samples. Image preprocessing involves resizing, HSV conversion, noise reduction through morphological operations, and feature extraction based on color and shape characteristics. The extracted features are used to train and test an ELM model. To improve classification performance and address potential overfitting in traditional ELM, this study introduces a new development called the Extended Extreme Learning Machine (EELM). The key innovation lies in modifying the calculation of the output weights βj, where a regularization term is introduced using ridge regression to stabilize learning and improve generalization. Experimental results show that the proposed system achieves 100% accuracy on the training data and an average accuracy of 83.3% on the testing data. The system also demonstrates robustness in handling varying lighting conditions and fruit shapes. These improvements enable EELM to better handle noisy or complex data by preventing over-reliance on randomly initialized hidden layer parameters. Consequently, EELM demonstrates improved reliability, making it more suitable for deployment in resourceconstrained real-world environments such as mobile or embedded systems.]]>
