MobileNetV2-D and multiple cameras for swiftlet nest classification based on feather intensity

Denny Indrajaya (Satya Wacana Christian University)
Hanna Arini Parhusip (Satya Wacana Christian University)
Suryasatriya Trihandaru (Satya Wacana Christian University)
Djoko Hartanto (PT Waleta Asia Jaya)

Indonesian Journal of Electrical Engineering and Computer Science
Vol 34, No 2: May 2024

MobileNetV2-D is a modified version of MobileNetV2, which is the novelty of this article. The algorithm is used to classify swiftlet nests into seven classes. In 2023, PT Waleta Asia Jaya is required to achieve a 7-fold increase in the export quota of swiftlet nests. To meet the quota, the company made a machine that can recognize swiftlet nest objects, which are classified into seven classes based on feather intensity, namely BRS, BR, BST, BS, BBT, BB, and BB2 for the light feathers to the heavy feathers, respectively. The input image is a combination of four images from four cameras with different positions, which adds to the novelty of MobileNetV2-D for the particular problem here. From the evaluation that has been carried out, the accuracy value of the MobileNetV2-D model was better than the MobileNetV2 model, i.e., the accuracy value of the MobileNetV2-D model was 99.9928% for the training dataset and 94.0723% for the testing dataset. Moreover, the speed of MobileNetV2-D is better than MobileNetV2- architecture.