Detecting and Extracting Cocoa Pods in the Natural Environment Using Deep Learning Methods

International audience Applications for image processing increasingly rely on object extraction and detection. The harvesting of cocoa pods produces the cocoa beans used to make chocolate. One of the most well-liked goods, chocolate, is made from these beans. It would be possible to develop intuitive ways to assess cocoa pods' ripeness, the presence of pod illnesses, or the amount of cocoa gathered from a specific location by separating cocoa pods from other natural materials. Our research intends to assess how well deep learning-based strategies work for locating and removing cocoa pods from their natural surroundings. This model will allow farmers to count cocoa pods on a specific cocoa plant and, generally, on a defined area. U-Net and FCN algorithms were employed. In the analysis of the results, the algorithms' validation phase yielded scores for U-Net and FCN of 93.61 and 93.06, respectively, and the test phase yielded scores for U-Net and FCN of 92.92 and 94.20, respectively. This indicates that FCN could generalize its learning more accurately than U-NET. These strategies were assessed using the Jaccard similarity coefficient and the Dice coefficient.