Food Waste Analysis through Computer Vision and Deep Learning

Project Objectives

Food Waste Analysis Using Deep Learning:
Utilize deep learning technologies to automatically analyze food waste, identifying the amount and types of leftovers.

Deep Learning Models and Cemera Used

- Intel Realsense Depth Camera D457
- Deeplabv3+
- Mask-RCNN
- YOLOv8

---

Scenario & Results

  1. Leftovers placed in the food disposal area are detected through a camera.
  2. Segmentation is applied to each identified leftover, and depth values are measured.
  3. The detected information is used to calculate and store the amount of food waste.

---

Label Image for Segmentation & Box Filtering

• The custom dataset was labeled using 'LabelMe'
• Apply label to the dataset images.

---

Deep Learning Model Training

The trained DeepLabV3 model with a ResNet101 backbone was ultimately selected for its visually detailed and successful segmentation.

- nvidea A100 GPU (from KWU)
- labels -> popcorn chicken, danmuji, rice, salad, donggeurangddaeng  
- deeplabv3_resnet101(backbone) -> 100 epoch  
- fcn_resnet50(backbone)        -> 100 epoch
- Mask-RCNN -> 100 epoch  

---

Segmentation Result & Box-filtered Result & Class Detection Result

• Results of Segmented Images and Box-Filtered Images.
• Class detection using segmentation and box-filtered images.

---

Integration of Trained Model with Realsense Camera

---

Calculation of Pixel Values for Each Type of Leftover

---

Retrieving and Calculating Depth Values from Realsense Camera

• Using the SIFT Algorithm to Extract the Position of a Tray and Retrieve Depth Information Based on Specific Locations on the Tray.
• Applying the average depth information to each side dish compartment to calculate the depth of the food in each compartment.
• Using the calculated depth values to later assign weights to the amount of each side dish.

---

Storing the Amount of Leftover Side Dishes in a CSV File

• Utilizing the Stored Data for Ingredient Recommendations and Customized Eating Habits

---

Prototype

---

Detailed Image of Results

---

Expected Outcomes

- Reduction of Food Waste  
- Provision of Customized Meal Plans  
- Improvement of Operational Efficiency in the Food Industry  

---