Defect detection is the most important step in the postpartum reprocessing of kiwifruit. However, there are some small defects difficult to detect. The accuracy and speed of existing detection algorithms are difficult to meet the requirements of real-time detection. For solving these problems, we developed a defect detection model based on YOLOv5, which is able to detect defects accurately and at a fast speed. The main contributions of this research are as follows: (1) a small object detection layer is added to improve the model’s ability to detect small defects; (2) we pay attention to the importance of different channels by embedding SELayer; (3) the loss function CIoU is introduced to make the regression more accurate; (4) under the prerequisite of no increase in training cost, we train our model based on transfer learning and use the CosineAnnealing algorithm to improve the effect. The results of the experiment show that the overall performance of the improved network YOLOv5-Ours is better than the original and mainstream detection algorithms. The mAP@0.5 of YOLOv5-Ours has reached 94.7%, which was an improvement of nearly 9%, compared to the original algorithm. Our model only takes 0.1 s to detect a single image, which proves the effectiveness of the model. Therefore, YOLOv5-Ours can well meet the requirements of real-time detection and provides a robust strategy for the kiwi flaw detection system.

The prevention and management of crop diseases play an important role in agricultural production, but there are many types of crop diseases and complex causes, and their prevention and identification add difficulties to the process. The traditional methods of identifying diseases mostly rely on human visual and manual inspection, which requires a certain amount of expert knowledge and experience. There are shortcomings such as strong subjectivity and low accuracy. This paper takes the common diseases of kiwifruit as the research object. Based on deep learning and computer vision models, and given the influence of a complex background in actual scenes on the detection of diseases, as well as the shape and size characteristics of diseases, an innovative method of target detection and semantic segmentation was proposed to identify diseases accurately. The main contributions of this research are as follows: We produced the world’s first high-quality dataset on kiwifruit. We used the target detection algorithm YOLOX, we stripped the kiwi leaves from the natural background and removed the influencing factors existing in the complex background. Based on the mainstream semantic segmentation networks UNet and DeepLabv3+, the experimental results showed that the ResNet101 network achieved the most effective results in the identification of kiwi diseases, with an accuracy rate of 96.6%. We used the training method of learning rate decay to further improve the training effect without increasing the training cost. After experimental verification, our two-stage disease detection algorithm had the advantages of high accuracy, strong robustness, and wide detection range, which provided a more efficient solution for solving the problem of precise monitoring of crop growth environment parameters.

Goat farming is one of the pillar industries for sustainable development of national economies in some countries and plays an active role in social and economic development. In order to realize the precision and intelligence of goat breeding, this paper describes an integrated goat detection and counting method based on deep learning. First, we constructed a new dataset of video images of goats for the object tracking task. Then, we took YOLOv5 as the baseline of the object detector and improved it using a series of advanced methods, including: using RandAugment to explore suitable data augmentation strategies in a real goat barn environment, using AF-FPN to improve the network’s ability to represent multi-scale objects, and using the Dynamic Head framework to unify the attention mechanism with the detector’s heads to improve its performance. The improved detector achieved 92.19% mAP, a significant improvement compared to the 84.26% mAP of the original YOLOv5. In addition, we also input the information obtained by the detector into DeepSORT for goat tracking and counting. The average overlap rate of our proposed method is 89.69%, which is significantly higher than the 82.78% of the original combination of YOLOv5 and DeepSORT. In order to avoid double counting as much as possible, goats were counted using the single-line counting based on the results of goat head tracking, which can support practical applications.

Disease diagnosis and control play important roles in agriculture and crop protection. Traditional methods of identifying plant disease rely primarily on human vision and manual inspection, which are subjective, have low accuracy, and make it difficult to estimate the situation in real time. At present, an intelligent detection technology based on computer vision is becoming an increasingly important tool used to monitor and control crop disease. However, the use of this technology often requires the collection of a substantial amount of specialized data in advance. Due to the seasonality and uncertainty of many crop pathogeneses, as well as some rare diseases or rare species, such data requirements are difficult to meet, leading to difficulties in achieving high levels of detection accuracy. Here, we use kiwifruit trunk bacterial canker (Pseudomonas syringae pv. actinidiae) as an example and propose a high-precision detection method to address the issue mentioned above. We introduce a lightweight and efficient image generative model capable of generating realistic and diverse images of kiwifruit trunk disease and expanding the original dataset. We also utilize the YOLOv8 model to perform disease detection; this model demonstrates real-time detection capability, taking only 0.01 s per image. The specific contributions of this study are as follows: (1) a depth-wise separable convolution is utilized to replace part of ordinary convolutions and introduce noise to improve the diversity of the generated images; (2) we propose the GASLE module by embedding a GAM, adjust the importance of different channels, and reduce the loss of spatial information; (3) we use an AdaMod optimizer to increase the convergence of the network; and (4) we select a real-time YOLOv8 model to perform effect verification. The results of this experiment show that the Fréchet Inception Distance (FID) of the proposed generative model reaches 84.18, having a decrease of 41.23 compared to FastGAN and a decrease of 2.1 compared to ProjectedGAN. The mean Average Precision (mAP@0.5) on the YOLOv8 network reaches 87.17%, which is nearly 17% higher than that of the original algorithm. These results substantiate the effectiveness of our generative model, providing a robust strategy for image generation and disease detection in plant kingdoms.

Soybean insect pests can seriously affect soybean yield, so efficient and accurate detection of soybean insect pests is crucial for soybean production. However, pest detection in complex environments suffers from the problems of small pest targets, large inter-class feature similarity, and background interference with feature extraction. To address the above problems, this study proposes the detection algorithm SP-YOLO for soybean pests based on YOLOv8n. The model utilizes FasterNet to replace the backbone of YOLOv8n, which reduces redundant features and improves the model’s ability to extract effective features. Second, we propose the PConvGLU architecture, which enhances the capture and representation of image details while reducing computation and memory requirements. In addition, this study proposes a lightweight shared detection header, which enables the model parameter amount computation to be reduced and the model accuracy to be further improved by shared convolution and GroupNorm. The improved model achieves 80.8% precision, 66.4% recall, and 73% average precision, which is 6%, 5.4%, and 5.2%, respectively, compared to YOLOv8n. The FPS reaches 256.4, and the final model size is only 6.2 M, while the number of computational quantities of covariates is basically comparable to that of the original model. The detection capability of SP-YOLO is significantly enhanced compared to that of the existing methods, which provides a good solution for soybean pest detection. SP-YOLO provides an effective technical support for soybean pest detection.

The neural network positioning algorithm is an algorithm that uses the principle of electromagnetic induction to activate a short-range wireless tag wirelessly to achieve information reading. It has the advantages of small size, low cost, and reusability. This paper compares several neural network structures of DNN, CNN, and RNN, and selects BP neural network to optimize its training through theory and practice, combining the data to compare their average error and average time used in indoor positioning practice, the average error of the optimized BP neural network in this article is smaller, and the positioning time is shorter, and combined with the practical data to obtain the efficient use of the division value n, which meets the high-precision requirements of indoor positioning and is more convenient for practical applications.