ChickenNet - an end-to-end approach for plumage condition assessment of laying hens in commercial farms using computer vision

Regular plumage condition assessment in laying hens is essential to monitor the hens’ welfare status and detect occurrence of feather pecking activities. However, commercial farms this a labor-intensive, manual task. This study proposes novel approach for automated using computer vision deep learning. It presents ChickenNet, an end-to-end convolutional neural network that detects simultaneously predicts score each detected hen. To investigate effect input image characteristics, method was evaluated images with without depth information resolutions 384 × 384, 512 512, 896 1216 pixels. Further, determine impact subjective human annotations, predictions were compared assessments one observer matching annotations two observers. Among all tested settings, performance metrics based on observers equal or better than ones single observer. The best result obtained among configurations mean average precision (mAP) 98.02% hen detection while 91.83% scores predicted correctly. Moreover, it revealed ChickenNet not generally enhanced by information. Increasing improved up resolution pixels, high accuracies (mAP > 0.96) could already be achieved lower resolutions. results indicate provides sufficient basis monitoring conditions farms.