RESUMO
Four deep learning frameworks consisting of Yolov5m and Yolov5m combined with ResNet50, ResNet-101, and EfficientNet-B0, respectively, are proposed for classifying tomato fruit on the vine into three categories: ripe, immature, and damaged. For a training dataset consisting of 4500 images and a training process with 200 epochs, a batch size of 128, and an image size of 224 × 224 pixels, the prediction accuracy for ripe and immature tomatoes is found to be 100% when combining Yolo5m with ResNet-101. Meanwhile, the prediction accuracy for damaged tomatoes is 94% when using Yolo5m with the Efficient-B0 model. The ResNet-50, EfficientNet-B0, Yolov5m, and ResNet-101 networks have testing accuracies of 98%, 98%, 97%, and 97%, respectively. Thus, all four frameworks have the potential for tomato fruit classification in automated tomato fruit harvesting applications in agriculture.
RESUMO
The strawberry (Fragaria × ananassa Duch.) is a high-value crop with an annual cultivated area of ~500 ha in Taiwan. Over 90% of strawberry cultivation is in Miaoli County. Unfortunately, various diseases significantly decrease strawberry production. The leaf and fruit disease became an epidemic in 1986. From 2010 to 2016, anthracnose crown rot caused the loss of 30-40% of seedlings and ~20% of plants after transplanting. The automation of agriculture and image recognition techniques are indispensable for detecting strawberry diseases. We developed an image recognition technique for the detection of strawberry diseases using a convolutional neural network (CNN) model. CNN is a powerful deep learning approach that has been used to enhance image recognition. In the proposed technique, two different datasets containing the original and feature images are used for detecting the following strawberry diseases-leaf blight, gray mold, and powdery mildew. Specifically, leaf blight may affect the crown, leaf, and fruit and show different symptoms. By using the ResNet50 model with a training period of 20 epochs for 1306 feature images, the proposed CNN model achieves a classification accuracy rate of 100% for leaf blight cases affecting the crown, leaf, and fruit; 98% for gray mold cases, and 98% for powdery mildew cases. In 20 epochs, the accuracy rate of 99.60% obtained from the feature image dataset was higher than that of 1.53% obtained from the original one. This proposed model provides a simple, reliable, and cost-effective technique for detecting strawberry diseases.
RESUMO
The application of optical absorption spectra in prognostic prediction has hardly been investigated. We developed and evaluated a novel two dimensional absorption spectrum measurement system (TDAS) for use in early diagnosis, evaluating response to chemoradiation, and making prognostic prediction. The absorption spectra of 120 sets of normal and tumor tissues from esophageal cancer patients were analyzed with TDAS ex-vivo. We demonstrated the cancerous tissue, the tissue from patients with a poor concurrent chemoradiotherapy (CCRT) response, and the tissue from patients with an early disease progression each had a readily identifiable common spectral signature. Principal component analysis (PCA) classified tissue spectra into distinct groups, demonstrating the feasibility of using absorption spectra in differentiating normal and tumor tissues, and in predicting CCRT response, poor survival and tumor recurrence (efficiencies of 75%, 100% and 85.7% respectively). Multivariate analysis revealed that patients identified as having poor-response, poor-survival and recurrence spectral signatures were correlated with increased risk of poor response to CCRT (P = 0.012), increased risk of death (P = 0.111) and increased risk of recurrence (P = 0.030) respectively. Our findings suggest that optical absorption microscopy has great potential to be a useful tool for pre-operative diagnosis and prognostic prediction of esophageal cancer.