<![CDATA[Fish quality control in traditional Indonesian markets remains a serious challenge because inspection relies almost entirely on manual visual assessment by a limited number of field officers who cannot cover all locations simultaneously. The problem is compounded when vendors manipulate spoiled fish with artificial dyes on the gills, making visual inspection unreliable. This study presents an integrated fish quality monitoring system combining a Convolutional Neural Network (CNN) model using MobileNetV2 architecture with a portable Internet of Things (IoT) device based on the ESP32 microcontroller. The system operates through a crowdsourcing mechanism where community members submit fish photos through a React.js web platform for automatic AI classification. When unfit-fish reports from one location reach a predefined threshold, an Early Warning System (EWS) automatically sends WhatsApp alerts to Fishery Department officers and activates physical alarms at the department office. Field officers then validate findings with a portable IoT device equipped with an MQ-135 ammonia gas sensor compensated by a DHT22 environmental sensor. The CNN model achieved an overall accuracy of 92.9%, precision of 91.9%, recall of 93.8%, and F1-score of 92.8% on a 300-image test set collected from Manado fish markets. Sensor testing across 25 samples showed fresh fish registering below 50 PPM and spoiled fish exceeding 100 PPM, including chemically manipulated samples that appeared visually fresh. WhatsApp alert dispatch averaged 1.7 ± 0.3 seconds after threshold was reached. These results confirm that dual-modality integration effectively bridges large-scale public reporting with objective chemical field validation.]]>
