The appearance quality of apples directly affects their price. To realize apple grading automatically, it is necessary to find an effective method for detecting apple surface defects. Aiming at the problem of a low recognition rate in apple surface defect detection under small sample conditions, we designed an apple surface defect detection network (ASDINet) suitable for small sample learning. The self-developed apple sorting system collected RGB images of 50 apple samples for model verification, including non-defective and defective apples (rot, disease, lacerations, and mechanical damage). First, a segmentation network (AU-Net) with a stronger ability to capture small details was designed, and a Dep-conv module that could expand the feature capacity of the receptive field was inserted in its down-sampling path. Among them, the number of convolutional layers in the single-layer convolutional module was positively correlated with the network depth. Next, to achieve real-time segmentation, we replaced the flooding of feature maps with mask output in the 13th layer of the network. Finally, we designed a global decision module (GDM) with global properties, which inserted the global spatial domain attention mechanism (GSAM) and performed fast prediction on abnormal images through the input of masks. In the comparison experiment with state-of-the-art models, our network achieved an AP of 98.8%, and a 97.75% F1-score, which were higher than those of most of the state-of-the-art networks; the detection speed reached 39ms per frame, achieving accuracy-easy deployment and substantial trade-offs that are in line with actual production needs. In the data sensitivity experiment, the ASDINet achieved results that met the production needs under the training of 42 defective pictures. In addition, we also discussed the effect of the ASDINet in actual production, and the test results showed that our proposed network demonstrated excellent performance consistent with the theory in actual production.
Background As one of the most widely planted fruit trees in southern China, citrus occupies an important position in the agriculture field and forestry economy in China. There are many kinds of citrus diseases. If citrus infected with diseases cannot be controlled in time, it easily seriously affects citrus production and causes large economic losses. Timely monitoring of disease characteristics in the citrus growth process is important for implementing timely control measures. Citrus images are easily disturbed by environmental factors such as dust, low light, clouds or leaf shadows. This makes it easy for some disease spot features in citrus pictures to be obscured. Occluded lesions cannot be effectively extracted and recognized. Second, similar characteristics of different diseases also make it difficult to distinguish the different types of diseases. However, the existing machine vision technology for identifying citrus diseases still has some difficulties in dealing with the above problems. Results This paper proposes a new citrus disease identification framework. First, a citrus image enhancement algorithm based on the MSR-AMSR algorithm is proposed, which can enhance the image and highlight the disease characteristic information. The AMSR algorithm can also greatly alleviate the interference of clouds and low light on image lesions, making the image features clearer. Second, an MF-RANet network is proposed to recognize citrus disease images. MF-RANet is composed of a main feature frame and a detail feature frame. The main feature frame uses the cross stacking structure of ResNet50 and RAM to extract the main features in the citrus image dataset. RAM is used to extract the attention weight in the feature layer, which enables RAM to give higher weight to disease features. The detailed feature frame path uses AugFPN to extract features from multiple scales and fuse the main feature frame path. AugFPN enables the network to retain more detailed features, so it can effectively distinguish similar features in different diseases. In addition, we use the ELU activation function not only to solve the problem of gradient explosion and gradient disappearance but also to effectively use the negative input of the network. Finally, we use the label smoothing regularization method to prevent overfitting the network in the classification process. Finally, the experimental results show that the highest detection accuracy of the network for Huanglong disease, Corynespora blight of citrus, fat spot macular disease, citrus scab, citrus canker and healthy citrus is 96.77%, 96.22%, 95.96%, 95.93%, 94.04% and 97.55%, respectively. Conclusions The citrus disease algorithm based on AMSR and MF-RANet can effectively perform the disease detection function. It has a high recognition rate for different kinds of citrus diseases. With the addition of AMSR preprocessing, RAM, AugFPN, ELU activation function and other structures, the MF-RANet network performance improves.
The growth of strawberry plants is affected by a variety of strawberry leaf diseases. Yet, due to the complexity of these diseases' spots in terms of color and texture, their manual identification requires much time and energy. Developing a more efficient identification method could be imperative for improving the yield and quality of strawberry crops. To that end, here we proposed a detection framework for strawberry leaf diseases based on a dual-channel residual network with a multi-directional attention mechanism (MDAM-DRNet). (1) In order to fully extract the color features from images of diseased strawberry leaves, this paper constructed a color feature path at the front end of the network. The color feature information in the image was then extracted mainly through a color correlogram. (2) Likewise, to fully extract the texture features from images, a texture feature path at the front end of the network was built; it mainly extracts texture feature information by using an area compensation rotation invariant local binary pattern (ACRI-LBP). (3) To enhance the model's ability to extract detailed features, for the main frame, this paper proposed a multidirectional attention mechanism (MDAM). This MDAM can allocate weights in the horizontal, vertical, and diagonal directions, thereby reducing the loss of feature information. Finally, in order to solve the problems of gradient disappearance in the network, the ELU activation function was used in the main frame. Experiments were then carried out using a database we compiled. According to the results, the highest recognition accuracy by the network used in this paper for six types of strawberry leaf diseases and normal leaves is 95.79%, with an F1 score of 95.77%. This proves the introduced method is effective at detecting strawberry leaf diseases.
Accurate pear sorting plays a crucial role in ensuring the quality of pears and increasing the sales of them. In the domain of intelligent pear sorting, precise target detection of pears is imperative. However, practical implementation faces challenges in achieving adequate accuracy in pear target detection due to the limitations of computational resources in embedded devices and the occurrence of occlusion among pears. To solve this problem, we built an image acquisition system based on pear sorting equipment and created a pear dataset containing 34,598 pear images under laboratory conditions. The dataset was meticulously annotated using the LabelImg software, resulting in a total of 154,688 precise annotations for pears, pear stems, pear calyxes, and pear defects. Furthermore, we propose an Extremely Compressed Lightweight Model for Pear Object Detection (ECLPOD) based on YOLOv7’s pipeline to assist in the pear sorting task. Firstly, the Hierarchical Interactive Shrinking Network (HISNet) was proposed, which contributed to efficient feature extraction with a limited amount of computation and parameters. The Bulk Feature Pyramid (BFP) module was then proposed to enhance pear contour information extraction during feature fusion. Finally, the Accuracy Compensation Strategy (ACS) was proposed to improve the detection capability of the model, especially for identification of the calyces and stalks of pears. The experimental results indicate that the ECLPOD achieves 90.1% precision (P) and 85.52% mAP50 with only 0.58 million parameters and 1.3 GFLOPs of computation in the homemade pear dataset in this paper. Compared with YOLOv7, the number of parameters and the amount of computation for the ECLPOD are compressed to 1.5% and 1.3%, respectively. Compared with other mainstream methods, the ECLPOD achieves an optimal trade-off between accuracy and complexity. This suggests that the ECLPOD is superior to these existing approaches in the field of object detection for assisting pear sorting tasks with good potential for embedded device deployment.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.