Rancang Bangun Sistem Deteksi Label Kardus Berbasis Model Kecerdasan Buatan YOLO dan EasyOCR serta ESP32-CAM

The Design of Cardboard Box's Label Detection System Using Artificial Intelligence Models of YOLO and EasyOCR and ESP32-CAM

Sistem Deteksi Label Kardus berbasis model YOLO dan EasyOCR merupakan sistem yang dirancang untuk mendeteksi keberadaan kardus serta tulisan pada label kardus menggunakan model kecerdasan buatan. Masukkan gambar untuk sistem deteksi label berasal dari ESP32-CAM yang mengirimkan gambar ke skrip Python via protokol WiFi. Penelitian ini bertujuan untuk menciptakan sistem pendeteksi label kardus yang cepat serta memiliki akurasi yang tinggi. Sabuk konveyor dibangun untuk menciptakan suasana simulasi asli.  Adapun spesifikasi utama sabuk konveyor yaitu panjang total sabuk konveyor = 30 cm, dan kecepatan sabuk konveyor = 1,626 cm/s. Berdasarkan hasil penelitian dapat diketahui bahwa waktu rata-rata respon pemrosesan dari ESP32-CAM adalah 0,762 detik. Dalam 40 iterasi pelatihan model YOLO dengan set data kustom, model terbaik yang dihasilkan memiliki loss deteksi objek sebesar 0,00997 dan mean Average Precision (mAP) sebesar 0,983. Hasil pendeteksian objek kardus dengan model YOLO yang sudah dilatih menghasilkan rata rata nilai confidence sebesar 58,002% pada gambar kardus berlabel. Model YOLO menggunakan threshold sebesar 0,5 yang memiliki rasio akurasi:error sebesar  19:1 dan hasil deteksi objek non kardus sebanyak 2/20 atau 10%. Setelah model YOLO berhasil mendeteksi kardus berlabel, maka digunakan model EasyOCR untuk membaca kata pada label kardus. Rata-rata nilai confidence tertinggi dicatatkan oleh label “jeruk” dengan nilai confidence sebesar 0,973364, sedangkan waktu pemrosesan gambar tercepat dicatatkan oleh label “ayam” dengan waktu rata-rata pemrosesan sebesar 11,47 detik. Hasil pembacaan label, nilai confidence pendeteksian kardus dan pembacaan label kemudian ditampilkan melalui laman website sederhana untuk mempermudah pengawasan secara real-time.

Kata Kunci: Sistem Deteksi Label Kardus, Kecerdasan Buatan, ESP32-CAM.

Cardboard Box’s Label Detection System based on YOLO and EasyOCR models is a system that is built to detect a cardboard presence and its label using Artificial Intelligence. Image input for the label detection system were fed through an ESP32-CAM camera that are continuously sending images to a Python script using the WiFi protocol. The purpose of this research is to come up with a cardboard’s label detection system that are both fast, and highly accurate. Conveyor belt was also built to help the simulation. The conveyor belt’s main specifications are its full length of 30 cm and its velocity of 1,626 cm/s. Based on the research conducted it is known that the average image processing time for an ESP32-CAM is 0,762 seconds. The characteristics of the best YOLO model has a detection loss value of 0,00997 and mean Average Precision (mAP) value of 0,983 after 40 epochs of training the YOLO model with custom dataset. The trained custom YOLO model were then being used to detect the cardboard boxes, which resulted in 58,002% of average confidence value across all images of cardboard box with label on top of it. YOLO model used a 0,5 threshold that has a ratio of accuracy:error of 19:1 and a non-cardboard box detection of 2/20 or 10%. After the success of detecting the cardboard boxes with the YOLO model, EasyOCR model was used to read the label. The highest average confidence value was achieved by the “jeruk” label with confidence value of 0,973364, while the fastest average image processing time was achieved by the “sawi” label with average processing time of 7,249 seconds. The results of label reading, confidence value of cardboard box detection and label reading are then displayed in a simple website for easier real time monitoring.

Keywords: Cardboard Box’s Label Detection System, Artificial Intelligence, ESP32-CAM